The use of phosphoproteomic data to identify altered and signaling pathways in cancer

By

Sara Renee Savage

Thesis

Submitted to the Faculty of the

Graduate School of Vanderbilt University

in partial fulfillment of the requirements

for the degree of

MASTER OF SCIENCE

in

Biomedical Informatics

August 10, 2018

Nashville, Tennessee

Approved:

Bing Zhang, Ph.D.

Carlos Lopez, Ph.D.

Qi Liu, Ph.D.

ACKNOWLEDGEMENTS The work presented in this thesis would not have been possible without the funding provided by the NLM training grant (T15-LM007450) and the support of the Biomedical Informatics department at Vanderbilt. I am particularly indebted to Rischelle Jenkins, who helped me solve all administrative issues. Furthermore, this work is the result of a collaboration between all members of the Zhang lab and the larger CPTAC consortium. I would like to thank the other CPTAC centers for processing the data, and Chen Huang and Suhas Vasaikar in the Zhang lab for analyzing the colon cancer copy number and proteomic data, respectively. All members of the Zhang lab have been extremely helpful in answering any questions I had and offering suggestions on my work. Finally, I would like to acknowledge my mentor, Bing Zhang. I am extremely grateful for his guidance and for giving me the opportunity to work on these projects.

ii TABLE OF CONTENTS

Page

ACKNOWLEDGEMENTS ...... ii

LIST OF TABLES...... v

LIST OF FIGURES ...... vi

LIST OF ABBREVIATIONS ...... viii

Chapter

1. Introduction ...... 1

Dysregulation of Cellular Components of Cancer ...... 1 Multi-Omics Integration in Cancer ...... 2 Kinases and in Cancer ...... 3 The Use of Phosphoproteomic Data to Study Signaling ...... 3

2. Overview of Resources for Studying Kinases, Phosphatases, and Phosphorylation Sites: Tools for Analyzing Phosphoproteomic Data ...... 5

Introduction ...... 5 Methods ...... 6 Results ...... 8 Discussion ...... 23

3. Proteomic Landscape of Kinase Signaling in Cancer ...... 26

Introduction ...... 26 Methods ...... 27 Results ...... 29 Discussion ...... 36

4. Characterization of Molecular Subtype-Specific Driver Subnetworks in Breast Cancer ...... 38

Introduction ...... 38 Methods ...... 39 Results ...... 40 Discussion ...... 47

iii 5. Phosphoproteomic Data Reveal a Dual Role for RB1 in Colon Cancer...... 48

Introduction ...... 48 Methods ...... 49 Results ...... 51 Discussion ...... 59

6. Conclusions and future directions...... 61

Summary ...... 61 Evaluation of Technology...... 61 Kinase Signaling in Cancer ...... 62 Kinase Signaling Comparison Among Tissues ...... 62

Appendix

A. ROC curve for substrate prediction of PKC ...... 63

B. Website URLs for bioinformatics tools ...... 64

C. Breast cancer subtype driver subnetworks ...... 67

D. WikiPathways enrichment results for subtype driver subnetworks ...... 71

E. GO BP enrichment for with altered phosphorylation in colon cancer ...... 74

F. Colon cancer-associated phosphorylation sites ...... 75

REFERENCES ...... 77

iv LIST OF TABLES

Table Page

1. Knowledge bases of human kinases and phosphatases ...... 9

2. Databases of phosphorylation sites ...... 10

3. Available phosphorylation site and kinase-substrate prediction tools ...... 14

4. Kinase activity prediction and phosphoproteomic dataset analysis tools ...... 18

5. Resources for studying the effect of mutations on kinases and phosphorylation

sites ...... 21

6. Kinase-inhibitor relationship resources ...... 22

7. Miscellaneous kinase signaling tools ...... 22

8. Number of identified proteins in three cancer datasets ...... 29

9. Predicted proteogenomic mutation effect from ReKINect ...... 36

10. Colon phosphoproteomic data by the numbers ...... 51

v LIST OF FIGURES

Figure Page

1. Network of phosphorylation site and kinase-substrate interaction databases ...... 11

2. Number of substrates per kinase and ...... 13

3. Network of phosphorylation site predictor tools and the resources used to make

predictions ...... 16

4. ROC curves for substrate prediction of four kinases ...... 17

5. True and false positive predictions for kinase activity prediction tools ...... 19

6. Kinase activity AUC for the GSEA and z score methods using various substrate

sets ...... 20

7. Intersection of detected kinases and phosphatases in three different cancer

types ...... 29

8. in CPTAC and HPA data ...... 30

9. Number of amino acids in proteins that were and were not detected by mass

spectrometry ...... 31

10. Log2 mRNA expression of proteins that were and were not detected by mass

spectrometry ...... 32

11. Heatmap of inferred kinase and phosphatase activity scores for breast cancer

samples ...... 34

12. Multi-omics profile of MERTK and NT5C in breast cancer ...... 35

13. Venn diagram of significant nodes in subtype-specific driver subnetworks ...... 40

14. Relative differential phosphorylation abundance in breast cancer subtypes ...... 42

vi 15. Overlap of the basal driver subnetwork and the DNA damage response pathway .. 43

16. Nodes unique to the basal subnetwork...... 44

17. Nodes unique to the luminal A subnetwork ...... 45

18. Relative phosphorylation of substrates of kinases ...... 46

19. Difference between phosphorylation in colon tumor and normal ...... 52

20. Characterization of phosphorylation sites in colon cancer ...... 53

21. Overlap of cancer-associated and colon cancer-associated proteins and

phosphoproteins ...... 54

22. Kinase activity in colon tumor compared to normal ...... 55

23. RB1 characteristics in colon cancer ...... 56

24. RB1 correlation with proliferation and apoptosis markers...... 57

25. RB1 effect on cell line sensitivity to CDK inhibitors ...... 58

26. Summary of RB1 activity in colon cancer ...... 59

vii LIST OF ABBREVIATIONS

ANN artificial neural network AUC area under the curve CGC Cancer Census CPTAC Clinical Proteomic Tumor Analysis Consortium DEPOD DEPhOsphorylation Database DL downloadable EKPD Eukaryotic Kinase & Database EMT epithelial-to-mesenchymal transition ERa estrogen receptor alpha ERb estrogen receptor beta FDR false discovery rate GO BP Biological Process GDSC Genomics of Drug Sensitivity in Cancer GSEA gene set enrichment analysis HINT High-quality INTeractomes HMM hidden Markov model HPA Human Protein Atlas HPRD the Human Protein Reference Database HT high-throughput ICGC International Cancer Genome Consortium IMAC immobilized metal affinity chromatography iTRAQ isobaric tags for relative and absolute quantification KEA2 Kinase Enrichment Analysis 2 KSD Kinase Sequence Database LC liquid chromatography LKB1 liver kinase B1 lncRNA long non-coding RNA LT low-throughput MAP4 associated protein 4 MAPK mitogen activated protein kinase MCLP MD Anderson Cell Lines Project MELK maternal embryonic leucine zipper kinase METABRIC Molecular Taxonomy of Breast Cancer International Consortium MIMP Mutations Impact on Phosphorylation MOAC metal oxide affinity chromatography MS mass spectrometry MS/MS tandem mass spectrometry NES normalized enrichment score

viii PCA principal component analysis PDK1 pyruvate dehydrogenase kinase PDPK1 3-phosphoinositide-dependent protein kinase 1 PEG3 paternally expressed gene 3 PI3K phosphatidylinositol-3-kinase PPI protein-protein interaction PSSM position specific scoring matrices PTM post-translational modifications RB1 retinoblastoma 1 ROC receiver operating characteristic RPPA reverse phase protein array RWR random walk with restart SAMHD1 sterile alpha motif and HD domain containing protein 1 SNV single nucleotide variation ssGSEA single sample gene set enrichment analysis SVM support vector machine TCGA The Cancer Genome Atlas TMT tandem mass tag TOP2A topoisomerase II alpha TUBA1B tubulin alpha-1B chain UNSP unspecified

ix CHAPTER 1

INTRODUCTION

Dysregulation of Cellular Components in Cancer Despite more than 2000 years of research on cancer, we still do not fully understand the disease1. Cancer is a complicated process that can occur in nearly every cell type and is characterized by abnormal cell proliferation. It is the second leading cause of death in the United States; more than half a million people die each year due to cancer2. Cancer is a genetic disease driven by changes in DNA3. Genetic mutations include single nucleotide polymorphisms, small insertions or deletions, chromosomal translocations, and copy-number alterations of large regions. Some of these mutations might occur in the germline, while the majority are somatic. Germline mutations, which can be inherited and passed on to future generations, appear in most cells of the body. Hereditary cancer accounts for 5 to 10% of all cancer diagnoses4. Somatic mutations, however, cannot be inherited and usually occur in a small population of cells. Together, genetic aberrations cause changes in other cell components, leading to pathology. The first step in cell physiology is the of genes to RNA. Transcriptional regulation is itself a complicated process involving specific DNA promoter sequences, transcription factors, and regulatory proteins and RNAs. Therefore, dysregulation of transcription is one of the contributing factors to cancer pathology. Both the expression and regulation of genes can be affected5. For example, MYC is a transcription factor frequently mutated in cancer and these mutations often affect its function or prevent its degradation6. Furthermore, overexpression of MYC causes changes in the expression levels of numerous other genes and results in tumor growth7. In T-cell acute lymphoblastic , chromosomal translocation moves a set of genes closer to a DNA regulatory element, leading to increased expression8. In addition to coding RNAs, non-coding RNAs can be dysregulated in cancer. miRNAs are small non-coding RNAs responsible for finely regulating the expression of protein-coding mRNA. For example, amplification of the miR- 17-92 cluster of miRNAs allows for increased in lymphoma by inhibiting tumor suppressor translation9. Furthermore, long non-coding RNAs (lncRNAs) regulate through various mechanisms. In breast cancer, amplification of the lncRNA HOTAIR reduces tumor suppressor expression10. After RNA is transcribed, it is further translated into proteins. Protein translation is another point of dysregulation in cancer. First, abnormalities from previous steps can be transferred to proteins. Changes in mRNA expression usually lead to corresponding changes in protein levels. DNA mutations can be translated into the protein sequence, leading to changes in protein structure and function. For example, single missense mutations in the TP53 transcription factor decrease its ability to bind DNA11. Additionally, chromosomal translocation events can create new fusion proteins. A fusion of two kinases, BCR and ABL, forms after a translocation event between 9 and 22. The fusion affects regulation of these kinases, leading to a constitutively active enzyme12. Finally, the actual process of translation might contribute to cancer

1 development. The ribosome might have altered structure or function and altered signaling might affect the rate of translation13. Finally, all components of the cell can be further regulated by the addition of other molecules. DNA methylation influences gene transcription. Hypermethylation of some promoters in colon cancer inhibits tumor suppressor expression14. Proteins are also modified in several ways. Glycosylation, the addition of a carbohydrate to a protein, is altered in cancer. Tumors have overall changes in glycosylation compared to normal15. One effect is the disruption of E-cadherin junctions that contributes to metastasis16. Similarly, phosphorylation, the addition of a phosphate group, is dysregulated in cancer. Tumor cells have a different phosphorylation pattern than normal cells17. Mutations can also create or destroy phosphorylation sites, leading to altered signaling18. The altered cellular components are intricately connected and combine to produce cell phenotypes. While elucidating the effect of a single mutation on an individual cell is relatively easy, determining the result of the interplay between several mutations is much harder. Average solid tumors contain between 33 and 66 genes with single-base substitutions or small insertions or deletions19. They also have dozens of chromosomal translocations, as well as amplification of 17% of the genome and deletion of a further 16%20,21. Only a fraction of these mutations drive disease pathology, while the remaining are passenger mutations that do not contribute to tumor growth. Determining the driver mutations and how they contribute to cancer development and progression is essential for effective treatment. Studying these data together produce a much better overall picture of the cancer.

Multi-Omics Integration in Cancer The development of high-throughput methods has allowed the collection of many data types, resulting in compilations of large amounts of information about specific tumors and cell lines. Each cellular component can be assayed. The whole genome can be sequenced to find mutations (genomics). RNA sequencing determines expression of coding and non-coding transcripts (transcriptomics). Copy number alterations are often identified by gene arrays. Methylation is further assayed using tiling microarrays or BeadChips. Protein levels and post-translational modifications are evaluated using mass spectrometry or protein arrays (proteomics). Finally, clinical data can be collected for each patient and phenotype data can be collected for each cell line. There are a few large-scale studies collecting and integrating multi-omics information on cancer patients. The Cancer Genome Atlas (TCGA) was a large program that collected clinical, genomic, transcriptomic, and proteomic data for over 11,000 patients and 30 tumor types21. The Clinical Proteomic Tumor Analysis Consortium (CPTAC) extended this data to include mass spectrometry proteomic and phosphoproteomic data for three of the cancer types22. The Molecular Taxonomy of Breast Cancer International Consortium (METABRIC) collected similar data for over 2000 breast cancer patients23. Finally, the International Cancer Genome Consortium (ICGC) also collected data from several different countries on 21 tumor types24. Much work has been done creating bioinformatics tools and resources to integrate and analyze these diverse data types.

2 Kinases and Phosphatases in Cancer While many aspects of the entire cell system are involved in cancer, kinase signaling is an important component. Kinases, the second largest protein class, are enzymes that catalyze the transfer of a phosphate group from energy molecules to other substrates. The majority of kinases phosphorylate proteins, but other substrates include carbohydrates, amino acids, and lipids25. Phosphatases are the enzymes that reverse this process. This signaling process contributes to all of the essential cell pathways that are also hallmarks of cancer including metabolism, motility, proliferation, and the DNA damage response. These two groups, along with their substrates, are known to be dysregulated in cancer. For example, PTEN, a phosphatase and known tumor suppressor, is frequently mutated in cancer. The mutations inactivate PTEN, leading to constitutive activation of AKT kinase signaling26. The phosphatidylinositol-3-kinase (PI3K)/AKT pathway regulates cell survival in response to stress and promotes cell proliferation and migration. The gene for the driving kinase of this pathway, PIK3CA, is also one of the most frequently mutated genes in cancer27. The activating mutations on PIK3CA increase proliferation and enhance AKT signaling28. Another kinase signaling pathway frequently dysregulated in cancer is the mitogen activated protein kinase (MAPK) pathway. The MAPK pathway is involved in many cell processes and can function to suppress or support tumor growth based on context29. In cancer, abnormal overexpression of the upstream receptor tyrosine kinases, SRC, or RAS upregulates MAPK signaling to increase proliferation and invasion30. Because kinases are such important targets, almost 40 kinase inhibitors have been approved by the FDA and are frequently used as anti-cancer agents31. Furthermore, some kinase inhibitors can be used for very specific mutations in personalized medicine. For example, vemurafenib specifically targets the kinase BRAF with the V600 mutation32. Imatinib targets the fusion kinase BCR-ABL, although it also inhibits the kinases c-KIT and PDGFR33. While kinase inhibitors have been instrumental in treating some patients, there are some downsides to using them. Because the active domains of many kinases are very similar, inhibitors frequently target many kinases34. Resistance to the drugs develops rapidly because of redundancy in the kinase signaling pathways35. Finally, toxicity to normal cells causes severe side effects for some inhibitors36.

The Use of Phosphoproteomic Data to Study Kinase Signaling Understanding which kinases are dysregulated in a patient and prioritizing kinases for drug development requires studying kinase signaling at a systems level. In the past, global kinase signaling dysregulation was primarily studied at the genomic level with determination of activating or inhibiting mutations. Individual kinases are studied in individual patients or cancer cell lines using molecular biology techniques. However, a method has recently been developed to study kinase signaling at the systems level for numerous cancer patients at once. Phosphorylated peptides in samples are enriched by affinity purification, optionally labeled for quantification, and identified by mass spectrometry. Because phosphorylation is the direct result of net kinase and phosphatase activity, it can be used as a read-out of active enzymes and their downstream pathways.

3 There are, however, numerous challenges with phosphoproteomic experiments. First, sample preparation and enrichment strategies can bias results. For example, the use of immobilized metal affinity chromatography to enrich phosphorylated peptides produces a higher proportion of multiply phosphorylated peptides than the use of metal oxide affinity chromatography37. Furthermore, phosphorylation is dynamic and unstable. When tumor samples are collected for phosphoproteomic analysis, cold ischemia time affects the results. Mertins et al found 24% of the phosphoproteome is regulated by cold ischemia, which complicates comparisons between patients if sample collection is not tightly controlled38. Finally, identification of the protein and the exact location of the phosphorylated residue is still a challenge39. Peptides can map to multiple proteins, which can affect site quantification. Additionally, spectra for phosphorylated sites within close proximity are hard to differentiate. In a study of 22 research groups analyzing the same phosphoproteomic data, the groups did not agree on site localization for 21% of the spectra40. Because mass spectrometry phosphoproteomics is a relatively new but promising technique, much work still needs to be done to understand its limitations and determine the best and most biologically relevant methods for analysis. This thesis aims to characterize the limitations of mass spectrometry in studying kinase signaling and demonstrate its utility in identifying new insights in cancer that cannot be determined using other methods.

4 CHAPTER 2

OVERVIEW OF RESOURCES FOR STUDYING KINASES, PHOSPHATASES, AND PHOSPHORYLATION SITES: TOOLS FOR ANALYZING PHOSPHOPROTEOMIC DATA

Introduction Post-translational modifications (PTMs) are an essential aspect of cell signaling. The enzymatic addition or removal of a molecule or chemical to a protein allows a cell to rapidly and reversibly respond to environmental stimuli. There are many types of PTMs, but phosphorylation, the method of transferring a phosphate group to a substrate, is the best-studied and is common in mammalian cells. Cells use phosphorylation as a molecular switch and carefully regulate the process. Furthermore, phosphorylation dysregulation has been hypothesized to contribute to diseases such as cancer17. In mammalian cells, proteins are phosphorylated primarily on serine, threonine, and tyrosine residues. However, phosphorylation occasionally occurs on other residues including histidine, as well as on non-protein substrates41. Kinases are enzymes responsible for catalyzing the transfer of the phosphate group to a substrate. There are over 500 human kinases, but the actual number varies among research groups due to differences in functional prediction and annotation of pseudogenes25,42. Phosphatases reverse the action of kinases by catalyzing the removal of the phosphate group. There are almost 200 human phosphatase genes and, like kinases, most phosphatases dephosphorylate protein substrates43. Phosphoproteomic data have emerged as a mechanism to study kinase signaling. To analyze this data, resources such as knowledge bases of kinases, phosphatases, and phosphorylation sites are required. Additionally, tools for prediction, visualization, and analysis are instrumental in understanding the data. While many tools have overlapping functions, they differ in underlying knowledge bases, algorithms, input and output format and data, accessibility, advantages, limitations, and maintenance. Additionally, a newly developed tool is usually compared to a similar, previously published tool, but comparisons often do not include real-world, biological use-cases. For example, the inference of kinase activity is a popular use for phosphoproteomic data. Methods used to infer activity include using permutation to determine non-uniform distribution of substrates, comparing mean phosphorylation of substrates using a Z-test, and assessing phosphorylation levels of regulatory sites on kinases44,45. There has been little validation of the methods and only one benchmarking paper study comparing a few of the methods has been published44. Finally, the targeted audience for many tools consists of biologists without computational backgrounds. However, biologists are rarely consulted for design input and never requested to test the final product. There is no comprehensive list of tools to aid those using phosphoproteomic data in their research. Therefore, this chapter aims to collect tools and resources that can be used to analyze phosphoproteomic data, perform some benchmarking comparisons to determine the best tool available, and assess usability of the tools from the standpoint of a user.

5 Methods Collection of Tools The OMICtools resource (https://omictools.com) is a manually curated collection of bioinformatics tools46. This site was searched in October 2017 for tools using the words ‘kinase’, ‘phosphorylation’, ‘phospho’, or ‘phosphatase’. In addition, several more tools were collected from the literature. Only tools that were freely available, still accessible, and non-obsolete were included. Tools specific for organisms other than human were discarded. The year of last update was assumed to be the year of publication unless otherwise noted on the website. The method of access can be by a website (Web) or by a downloadable, locally-run tool (Tool). DL indicates database information or tool results could be downloaded. The website URLs for all resources can be found in Appendix B.

Testing Knowledge Bases Each website was accessed in October 2017. A protein was submitted to the search function and the links provided in the results were tested. Data statistics were collected for human proteins from downloadable files where possible and from websites or manuscripts for online-only resources.

Identifying the Human and Phosphatome Human protein kinases were downloaded from KinBase25 (http://kinase.com/web/ current/) and non-protein kinases annotated with the term KW-0418 were downloaded from Swiss-Prot47 (http://www.uniprot.org) in January 2017. Human phosphatases, excluding inactive phosphatases and those with the designation ‘pseudophosphatase,’ were downloaded in April 2017 from the DEPhOsphorylation Database48 (DEPOD, http://depod.bioss.uni-freiburg.de/) and Phosphatome.Net43. All gene identifiers (HGNC, UniProt) were updated to the June 2017 versions.

Testing Kinase-Substrate Prediction Tools Tools predicting kinases for phosphorylation sites were accessed through local tool installation or through the tool’s website. PhoScan49 and phos_pred50 were run locally on a Windows laptop, while NetPhorest51, NetworKIN52, iGPS53, GPS54, PhosphoPredict55, and MusiteDeep56 were run locally on a Mac laptop. PhosphoPICK57, NetPhos58, Musite59, and pkaPS60 were accessed via their websites. Gold standard positive and negative phosphorylation sites for five kinases (CDK1, CK2, MAPK1, PKA, and PKC) were downloaded from dbPTM61. Positive sites were phosphorylation sites experimentally validated to be phosphorylated by a particular kinase. Negative sites were phosphorylatable residues not known to be phosphorylated on the same proteins. Only human serine and threonine sites were used. Tools were set with the lowest threshold if they did not have an option to return scores for all sites. For each site, the maximum score was retained if the tool predicted for more than one related kinase (e.g., the maximum score of PKCalpha and PKCbeta on the same site). If a tool did not return a score for a site, the lowest possible score was given to the site. The receiver operating characteristic (ROC) curve and area under the curve (AUC) were calculated for the results from each tool using the R package ROCR62.

6 Testing Kinase Activity Prediction Tools A phosphoproteomic dataset from a cell line experiment with 20 kinase inhibitors was used to test kinase activity prediction tools63. The R programming environment was used to create files in the input format for each tool. Significantly downregulated sites for each inhibitor were submitted to KEA264 and significantly inhibited kinases were defined as those with false discovery rate (FDR) < 0.05 and at least 3 overlapping substrates. The log2 fold change for each thirteenmer phosphorylation site (±6 amino acids surrounding the phosphorylated site) was submitted to PHOXTRACK65 (1000 permutations, minimum number of substrates = 3, weighted statistics). Significantly inhibited kinases were defined as those with FDR < 0.05 and normalized enrichment value < 0. The fold change for each site with each inhibitor was submitted to the KSEA app website45 and significantly inhibited kinases were defined as those with FDR < 0.05, at least 3 substrates in the dataset, and a z score < 0. The substrates of kinases from PhosphoSitePlus66 (version July 2017) and Signor67 (version October 2017) were used for IKAP68. IKAP was run locally on a Mac laptop with the bounds between -11 and 11 and 50 iterations. The 5 kinases with the lowest activity scores for each experiment were chosen. The positive set were kinases known to be inhibited by each drug (supplementary table in reference 63); all other kinases were considered to be negative. The significant kinases for each tool were counted for presence in the positive and negative sets.

Creating Gold Standard Sets for Testing Kinase Activity Inference Twenty-five known kinase targets of the 20 inhibitors used in the Wilkes’ experiment with at least six substrates in the data were chosen for analysis. The gold standard positive set was the 60 known inhibitor-target pairs. The 20 inhibitors were then paired with each of the twenty-five kinases that were not known targets of that inhibitor. The gold standard negative sets were created by randomly selecting 60 inhibitor-non- target pairs 20 different times.

Method Comparison for Kinase Activity Inference Pre-ranked gene set enrichment analysis (GSEA) was implemented using WebGestaltR69. The z score method from the KSEA app was also implemented in R. Briefly, the mean log2 fold change for all sites was subtracted from the mean log2 fold change for substrates in the set. This difference was multiplied by the square root of the number of sites in the substrate set and divided by the standard deviation across all sites in the dataset. Phosphorylation sites were defined as the thirteenmer peptide and the median log2 fold change was calculated for multiple peptides referring to the same site. Activity scores were calculated as the signed -log10(FDR). The sign followed the normalized enrichment score (NES) from GSEA or the z score from the z score method.

Kinase Activity Inference Using Substrate Sets with Different Experimental Evidence Experimentally validated substrate sets were created for the 25 kinases from PhosphoSitePlus (version May 2018) annotated as in vivo or in vitro experimental evidence. In silico substrate sets were generated from NetworKIN using the NetworKIN score ³ 2. Analysis was limited to the 16 kinases with at least 3 substrates in each set.

7 Kinase Activity Inference Using Substrate Sets from Different Databases Substrate sets were created for the 25 kinases from 5 different databases. Thirteenmer sequences for substrates from each database were created by mapping to their respective protein sequences. Substrates for PhosphoSitePlus (version May 2018) were combined with those from Signor (version May 2018), Swiss-Prot (version May 2018), the Human Protein Reference Database70 (HPRD, version 9), and Phospho.ELM71 (version 9.0). Two additional sets of a combination of PhosphoSitePlus + HPRD + Swiss- Prot and a combination of all five databases were created. Analysis was limited to the 23 kinases with at least 3 substrates in every database combination.

AUC for Kinase Activity Inference AUC was calculated using the ROCR R package for each of the 20 negative sets with the positive set. AUC values within a method were compared using ANOVA with a Tukey’s post hoc analysis for differences between pairs. AUC values between methods were compared by t-test. Significance was defined as p < 0.05.

Results Knowledge Bases of Kinases and Phosphatases Knowledge bases for kinase signaling can be separated into those collecting information on the enzymes, and those collecting experimentally validated phosphorylation sites. Of the 15 different resources that collect information specifically on protein kinases and phosphatases, 12 provide data on kinases, while 4 provide data on phosphatases (Table 1). Only one resource, the Eukaryotic Protein Kinase & Protein Phosphatase Database (EKPD) contains information on both types of enzymes72. The databases contain various types of data and some include an option for downloading files, while others are only available as an online website. The kinase knowledge bases can be further separated into two different types: those that include comprehensive data on all known protein kinases, and those that were developed for a specific purpose, such as collecting driver mutations in kinases. Notably, no kinase resource collects data on non-protein kinases. KinBase, which was developed by Gerard Manning, contains 538 protein kinases and is considered the primary source of human protein kinases and their classification25. Many other resources base their kinase list on KinBase. Kinomer, Kinase Sequence Database (KSD), and KinG are general kinase sequence databases that provide very little other information and are outdated73–75. KinMutBase, a collection of disease-causing mutations in protein kinase domains, is also outdated, contains data on only 31 kinases, and primarily consists of broken links76. KinWeb and EKPD provide gene and protein identifiers, classification, description, and sequence information, but these data can also be found in other resources. However, KinWeb does have prediction of the disulfite bonding state of cysteines in the protein, as well as prediction of alpha helices, and EKPD presents data in an easy-to-read format72,77. Use of the remaining general resources depends on which data you want to access. KinaseNET and ProKinO contain the most comprehensive databases on protein kinases, but both are only available as online resources78. They include protein

8 sequences, links to the kinases in other databases (e.g., UniProt, Ensembl, ), information on the kinase domains, expression in tissue, and disease associations. ProKinO specifically contains pathway information, mutations and their disease associations, chromosomal location of the kinase, and links to published manuscripts. KinaseNET includes PTMs, known binding partners, inhibitors, upstream kinases, downstream substrates, and information about regulation. KinaseNET provides all data on a single page, while ProKinO requires more than 10 clicks on separate tabs and pages to obtain all information on a kinase. For studying diseases, MOKCa and Kin-Driver specifically have data on protein kinase mutations79,80. MOKCa has tissue specificity of mutations while Kin-Driver focuses on driver mutations and reports whether the mutation is activating or inactivating. Finally, KLIFS provides structural information for approximately half of the protein kinases bound to various ligands81. Because phosphatases are less well studied than kinases, there are fewer resources dedicated to their collection. EKPD provides the same information for phosphatases as it does for kinases. HuPho, however, was the first comprehensive collection of phosphatases and the database includes pathway and substrate data, as well as siRNA phenotype data and links to orthologs in other species82. DEPOD used data from HuPho as a starting point and therefore contains much of the same information48. Finally, Phosphatome.Net is the phosphatase version of KinBase43. The website contains basic classification and sequence information.

Last Method of Name Version Enzyme Protein Number Reference Update Access KSD 2002 Web|DL Protein Kinases 913 74 KinWeb 2005 Web Protein Kinases 519 77 Kinomer 2008 Web|DL 1 Protein Kinases 505 73 MOKCa 2008 Web Protein Kinases 423 79 HuPho 2012 Web|DL Phosphatases 313 82 EKPD 2013 Web 1.1 Protein Kinases and Phosphatases 676 72 KinBase 2014 Web Protein Kinases 538 25 Kin-Driver 2014 Web|DL Protein Kinases 518 80 KinG 2014 Web Protein Kinases 1813* 75 KinMutBase 2015 Web|DL 4 Protein Kinases 31 76 DEPOD 2016 Web|DL 1.1 Phosphatases 239 48 ProKinO 2016 Web 2 Protein Kinases 538 78 KinaseNET 2017 Web Protein Kinases >530 Phosphatome 2017 Web 3 Phosphatases 189 43 KLIFS 2018 Web|DL 2.3 Protein Kinases 285 81,83

Table 1. Knowledge bases of human kinases and phosphatases. *Indicates the inclusion of an unknown number of non-human proteins.

Generating the List of Human Kinases and Phosphatases After evaluation of the knowledge bases, I collected the human protein kinases from KinBase because it contained the most comprehensive list. Because no resource includes non-protein kinases, these enzymes were collected from Swiss-Prot (http://www.uniprot.org) annotated with the term KW-0418 (‘Kinase’). Notably, other ontologies do not have a term to specify proteins with phosphotransferase activity. For example, the Gene Ontology term ‘protein kinase activity’, GO:0050222, contains 1246

9 human genes, many of which are regulators of kinase activity and do not contain phosphotransferase activity themselves. The final collection contained 687 unique human kinases. This corresponded to 688 genes, as CKMT1A and CKMT1B produce the same protein product despite being separate genes. Of these 688 genes, 538 encode protein kinases. Active human phosphatases were collected from DEPOD. Additional human phosphatases that were not in DEPOD and were not verified pseudophosphatases were added from Phosphatome.Net. This resulted in 255 human proteins annotated as phosphatases.

Knowledge Bases of Phosphorylation Sites Besides information about specific kinases and phosphatases, data on phosphorylation sites are important for studying the signaling process. Phosphorylation site databases collect information on the location of phosphorylated residues in proteins from experimental data. These experiments can be low-throughput or high-throughput. High-throughput phosphorylation site identifications are assigned by probability unlike the more stringent experimental validation in low-throughput experiments, but some databases combine sites from both types of experiments without identifying the source experiment type. In addition to phosphorylation site information, about 75% of the 23 resources collect interactions between kinases or phosphatases and their substrates (Table 2). These often do not include the exact phosphorylation site, but instead provide interactions between an enzyme and its substrate at the gene level.

Last Method of Name Version Sites Proteins Kinases Phosphatases Data Type Reference Update Access PhosphoPep 2007 Web|DL 2.0 3,980 MS 84,85 Phospho.ELM 2010 Web|DL 9.0 26,651 5,374 250 HT, LT 71,86,87 Phospho3D 2010 Web|DL 2.0 1,770 59 HT, LT 88 HPRD 2010 Web|DL 9 78,005 11,807 291 42 UNSP 70,89,90 PHOSIDA 2011 Web|DL 3.24 22,806 5,175 MS 91,92 PTMfunc 2012 Web 31,165 MS 93 HuPho 2012 Web|DL 190 121 55 UNSP 82 SubPhosDB 2012 Web|DL 1 137,153 17,297 238 UNSP 94 RegPhos 2013 Web|DL 2.0 66,301 10,849 380 UNSP 95,96 ANIA 2013 Web|DL 305 220 LT 97,98 PhosphoNetworks 2013 Web|DL 1,140 255 UNSP 99 Kinome NetworkX 2014 DL 173,460 18,610 357 UNSP 100 ProteomeScout 2014 Web|DL 2 290,007 23,387 MS 101,102 dbPTM 2015 Web|DL v2016 76,736 10,648 >360 UNSP 61,103,104 dbPAF 2016 Web|DL 1.0 244,034 18,773 UNSP 105 PhosphoAtlas 2016 DL 2,595 1,284 501 UNSP 17 DEPOD 2016 Web|DL 1.1 253 210 88 UNSP 48 KANPHOS 2016 Web β 73 MS 106 Phosphatome 2017 Web 3 6,008 2,000 319 106 UNSP 43 PhosphoNET 2017 Web 966,817† 22,698 488 UNSP+pred 107 PhosphoSitePlus 2018 Web|DL May-18 235,406 20,086 370 HT, LT 66 Swiss-Prot 2018 Web|DL May-18 40,150 7,971 350 UNSP 47,108 Signor 2018 Web|DL 2.0 4,354 1,605 415 98 UNSP 67,109

Table 2. Databases of phosphorylation sites. The number of unique kinases and phosphatases reported to phosphorylate sites in the database is included. For some databases, these numbers include enzyme groups in addition to individual enzymes. Data type indicates whether the data are from mass spectrometry

10 (MS) experiments, separated high-throughput (HT) and low-throughput (LT) experiments, or whether the database combines data from both HT and LT experiments without specifying (UNSP). †Indicates inclusion of predicted phosphorylation sites (pred).

The four main resources for phosphorylation sites curated data manually from the literature (Figure 1). HPRD and Swiss-Prot are general databases of all proteins47,70. The remaining two, PhosphoSitePlus and Phospho.ELM, specifically contain phosphorylation site information66,71. Both PhosphoSitePlus and Swiss-Prot are frequently updated, while HPRD and Phospho.ELM were last updated in 2010. All four of these databases also include kinase information for sites if known.

literature literature experiments literature literature literature literature literature experiments literature literature literature literature literature

PhosphoPep PHOSIDA HPRD Swiss-Prot PhosphoSitePlus Phospho.ELM PhosphoNetworks ANIA HuPho

experiments literature literature

PDB KANPHOS Signor PTMfunc

SubPhosDB SysPTM dbPTM KinomeNetworkX Phospho3D DEPOD

dbPAF

PhosphoPOINT PhosphoAtlas ProteomeScout RegPhos PhosphoBase literature Figure 1. Network of phosphorylation site and kinase-substrate interaction databases. Gray nodes indicate databases that are no longer accessible. Arrows point from the knowledge source to the collecting database. Arrows originating from the four most highly used databases are colored by source (green=HPRD, blue=Swiss-Prot, red=PhosphoSitePlus, purple=Phospho.ELM).

Other smaller databases were generated through manual curation or publication of a laboratory’s own phosphorylation site data. KANPHOS collects phosphorylation sites in neural signaling identified by high-throughput experiments106. PHOSIDA is another collection of data that were primarily produced in cell lines91. PhosphoPEP integrates mass spectrometry experiments from Cell Signaling Technology and their own laboratory84,85. PTMfunc collects mass spectrometry experiments and adds functional predictions from various tools for each site93. Signor extracts high quality signaling interactions from the literature67. Finally, ANIA and PhosphoNetworks curate the literature for a specific purpose. ANIA collects phosphorylation sites that serve as binding sites for 14-3-3 proteins, while PhosphoNetworks creates a kinase-substrate network curated from the literature and a protein microarray experiment97,99. The remaining resources integrate phosphorylation sites and kinase information from other databases (Figure 1). The database dbPAF collects phosphorylation sites from several databases105. ProteomeScout also collects phosphorylation sites from other databases along with literature-curated experiments and provides a tool for analyzing a user’s data101. The database dbPTM collects all PTMs and the responsible enzyme from several sources61. KinomeNetworkX, RegPhos, and PhosphoAtlas curate and integrate

11 data specifically to create kinase-substrate networks17,95,100. PhosphoNET is an online- only tool that includes predicted phosphorylation sites in addition to those with experimental evidence107. SubPhosDB annotates phosphorylated proteins with subcellular localization94. Finally, Phospho3D specifically collects phosphorylation sites with 3D structures88. Five databases collect information on phosphatase-substrate interactions. As mentioned, DEPOD, HuPho, and Phosphatome.Net all curate enzyme interactions from the literature. HPRD and Signor also collect some site-specific phosphatase information. Each database contains a different number of phosphorylation sites and enzyme- substrate relationships depending on the source and method of collection (Table 2). ProteomeScout, PhosphoSitePlus, and dbPAF contain the most number of experimentally validated, downloadable sites. The site numbers for these three databases include specific protein isoforms, as do several other resources. PhosphoAtlas contains substrates for the most number of individual kinases. Signor, Swiss-Prot, RegPhos, Phospho3D, dbPTM, and Phospho.ELM have substrates for individual kinases and kinase families. Finally, PhosphoSitePlus has substrates for some specific kinase isoforms. Besides the general phosphorylation site databases, other specialized databases have formed using phosphorylated protein information. For example, PepCyber: P~PPep is an online searchable database of proteins that interact with phosphorylated proteins110.

Errors in Substrate Databases Based on these databases, PhosphoSitePlus is the best resource for experimentally-identified phosphorylation sites and kinases for phosphorylation sites. PhosphoSitePlus is frequently updated and well-curated. The downstream integrating databases suffer from ID mapping errors. For example, in PhosphoAtlas there is an entry for PEG (paternally expressed gene 3) phosphorylating CDC25B. PEG is not a known kinase, but pEg3 kinase (also known as maternal embryonic leucine zipper kinase, MELK) is known to phosphorylate CDC25B111. Many of the downstream databases also have issues with PDPK1 and PDK1. The gene PDPK1, 3-phosphoinositide-dependent protein kinase 1, produces a protein known to the biological community as PDK1. However, there is an additional kinase, pyruvate dehydrogenase kinase, that is produced by the gene PDK1. Databases that try to integrate sites frequently attribute the substrates of PDPK1 to PDK1. Finally, integrating databases propagate errors from the original databases. For example, HPRD contains an entry for PTPN11 phosphorylating PTK2B although PTPN11 is a known phosphatase and not a kinase. The original manuscript connected to this entry confirmed that PTPN11 is a phosphatase and that it just binds to PTK2B at that particular site112. Databases that collect information from HPRD, such as RegPhos and PhosphoAtlas, include this incorrect entry for PTPN11.

Known Substrates of Kinases and Phosphatases The four main databases together produce 493 substrate sets of individual kinases and kinase families (Figure 2A). PhosphoSitePlus contains the most unique sites, while other databases contribute only a few additional sites per kinase. CSNK2A1 has the most number of substrates (584), while over half of the sets contain fewer than 10 substrates.

12 For substrates of phosphatases, DEPOD, HPRD, and Phosphatome.Net combined produce sets for 83 phosphatases. The most unique information comes from DEPOD and Phosphatome.Net. The number of known sites for each phosphatase is far fewer than that for kinases. PPP2CA has the most substrates (167), while 70% of the phosphatases have fewer than 10 substrates (Figure 2B).

A Swiss−Prot 500 HPRD Phospho.ELM PhosphoSitePlus 400 Overlap

300

200 Number of Substrates 100

0 IKK LYN ATR TTK CK1 LCK CK2 ATM RET FYN SYK PKB PKA CSK PKC MET HCK CDK SRC PIM1 JAK2 INSR PAK2 PAK1 PLK2 PLK3 ABL1 ULK1 STK4 AKT2 TBK1 AKT1 NEK2 CDK6 CDK9 CDK7 CDK4 CDK5 CDK2 CDK1 GSK3 SGK1 CDC7 GRK2 EGFR CHUK AMPK MAPK IRAK4 HIPK2 MTOR IKBKE IKBKB STK11 LRRK2 DYRK2 FGFR1 PDPK1 GSK3A CHEK2 PRKD1 CHEK1 GSK3B MAPK7 TRPM7 PRKG1 MAPK9 PRKCZ MAPK8 MAPK3 MAPK1 AURKA AURKB ROCK1 CAMK2 PRKCE PRKCB PRKCA PRKDC PRKCD PRKCQ PRKCG FAM20C DYRK1A MAP3K7 PRKAA1 MAPK14 CSNK2B CSNK1E CSNK1D PRKACA CAMK2A RPS6KA5 RPS6KB1 RPS6KA3 RPS6KA1 CSNK2A2 CSNK1A1 CSNK2A1 MAPKAPK2

B DEPOD 150 HPRD Phosphatome.Net Overlap

100

50 Number of Substrates

0 EYA2 EYA1 EYA3 FCP1 ACP5 ACP1 SSH2 SSH3 PDP1 PDP2 SSH1 PTEN ACPP PDXP RPAP2 SSU72 PTPRJ PTPN7 PTPN3 PTPN5 PHPT1 PTPN2 PTPN6 PTPN1 PTPRT PTPRF PPP6C PPP4C PPP5C DUSP2 PPM1F DUSP4 PTPRB DUSP1 PTPRK DUSP3 PTPRE PTPRA EPM2A PTPRR CDKN3 PPM1K PTPRD PPM1E PPM1B PTPRC PPM1A PTPRO PPM1H PTPRG PPM1D PPM1G PTP4A3 PTPN18 PTPN14 PTPN22 PTPN13 PTPN12 PTPN11 PTPRZ1 PHLPP1 PHLPP2 DUSP14 DUSP22 DUSP26 CDC25B CDC25A CDC14B CTDSP2 CDC14A CTDSP3 CTDSP1 PPP1CB PPP3CA PPP3CB PPP2CB PPP1CA PPP2CA CDC25C PPP1CC PPP3CG PPP1CG UBASH3B

Figure 2. Number of substrates per kinase and phosphatase. A) Number of substrates for the top 100 kinases in four databases. Substrates present in more than one database are colored black while the remaining sites are unique to each database. B) Number of substrates for each phosphatase in DEPOD (blue), HPRD (green), Phosphatome.Net (red), or in more than one database (black).

13 Phosphorylation Site Prediction Tools Despite decades of research, very few phosphorylation sites have known kinases or phosphatases. Of the sites in PhosphoSitePlus, only about 3% have an experimentally validated human kinase. Therefore, numerous tools have been developed to predict which sites in a protein can be phosphorylated and which kinases phosphorylate that given site. These prediction tools were developed using a variety of features and methods and have been reviewed elsewhere113,114. The early versions of phosphorylation site predictors were motif-based. They generated the frequency of amino acids surrounding a site and searched for that pattern in protein sequences. Later tools used more sophisticated methods such as support vector machines (SVM), random forest, Bayesian probability, and position specific scoring matrices (PSSM)50,115–117. Besides amino acid sequence, tools included a vast array of features such as the 3D structure of the phosphorylation site, disorder score, data, and co-expression of kinases and substrates118–120. Others, like NetworKIN and iGPS, used protein-protein interaction data to filter predictions53,121. Table 3 provides an overview of all currently available tools to predict phosphorylation sites or kinases for phosphorylation sites. While a few tools have been developed to predict sites for phosphatases, only NetPhorest and NetworKIN are still accessible121.

Last Kinases/ Tool Version Prediction Type Method Type Reference Update Phosphatases Disphos 2004 1.3 phosphorylation sites bagged logistic regression Web 118

PPSP 2006 1.06 phosphorylation sites of Bayesian decision theory 68 Web 116 kinases KinasePhos2.0 2007 2.0 phosphorylation sites of SVM 58 Web 115 kinases pkaPS 2007 phosphorylation sites of scoring function 1 Web|DL 60 PKA PhoScan 2008 phosphorylation sites of scoring function 48 Web|Tool 49 kinases Phos3D 2009 phosphorylation sites and SVM 5 Web 119 some kinase specificity Musite 2010 1 phosphorylation sites and SVM 13 Web|DL 59 some kinase specificity PHOSIDA 2011 3.24 phosphorylation S and T SVM Web 91 Predictor sites Predikin 2011 phosphorylation sites of PSSM any Web|DL 117 kinases GPS-Polo 2012 1.0 phosphorylation sites of group-based scoring 1 Web|Tool 122 Plk function PSSM iGPS 2012 1.0.1 phosphorylation sites of GPS with PPI 407 Tool 53 kinases in vivo HMMpTM 2013 phosphorylation sites of HMM 9 Web|DL 123 kinases and topology PKIS 2013 phosphorylation sites of SVM 56 Web|DL 124 kinases CEASAR 2013 kinases for known naïve Bayes 289 DL 120 phosphorylation sites GPS 2014 3.0 phosphorylation sites of group-based scoring 464 Web|DL|Tool 54 kinases function PSSM NetPhorest 2014 2.1 phosphorylation sites of ANN&PSSM 244 Web|DL|Tool 51,121 kinases NetworKIN 2014 3.0 phosphorylation sites of naïve Bayes with PPI 123 Web|DL|Tool 52,121 kinases in vivo

14 phos_pred 2014 predicts phosphorylation random forest 54 Tool† 50 sites for kinases PhosphoSVM 2014 phosphorylation sites SVM Web 125

PSEA 2014 phosphorylation sites of GSEA 33 Web 126 kinases Scansite 2015 4 kinase motifs in proteins PSSM 70 Web|DL 127 KSP-PUEL 2015 phosphorylation sites of SVM ensemble 2* Tool 128 kinases PhosphoPICK 2016 phosphorylation sites of Bayesian network 107 Web|DL 57 kinases PhosD 2016 kinase-substrate probabilistic model 399 DL 129 relationships PhosphoNET 2017 phosphorylation sites of PSSM 488 Web 107 kinases NetPhos 2017 3.1 phosphorylation sites and ANN 17 Web|Tool† 58,130 some kinase specificity PhosphoPredict 2017 phosphorylation sites of random forest 12 Web|DL|Tool 55 kinases MusiteDeep 2017 phosphorylation sites and deep learning 5 Tool† 56 some kinase specificity PhosPred-RF 2017 phosphorylation sites random forest Web 131

Table 3. Available phosphorylation site and kinase-substrate prediction tools. *Indicates number of trained kinases, but tool can be trained with others. †Indicates tool is not available for all three main operating systems (Linux, Mac, Windows). SVM – support vector machine, PSSM – position specific scoring matrix, GSEA – gene set enrichment analysis, ANN – artificial neural network, HMM – hidden Markov model, PPI – protein-protein interaction

Almost all phosphorylation site predictors were trained using data from Phospho.ELM (Figure 3). Swiss-Prot and PhosphoSitePlus were also heavily used resources. Notably, almost all tools were developed using experimentally verified substrate data as the training set. Therefore, the tools are only able to predict the responsible kinase if there is existing data for substrates of that kinase.

15 RegPhos MusiteDeep SwissProt Scansite

PhosphoNetworks HPM Ensembl experiments

PSEA PhosD Predikin Musite KinasePhos2 DISPHOS literature

PhosphoSVM

PhoScan

IntAct Interpro PhosPred-RF PhosphoELM HPRD DIP BioGRID PKIS PhosphoSitePlus iGPS MINT pkaPS SysPTM

PhosphoPep literature CEASAR experiments KSP-PUEL

PPSP literature

literature Phos3D GPS

experiment PDB PhosphoPICK PhosphoNET PhosphoPredict

PepCyber experiments HMMpTM NetPhos PHOSIDA_DB literature literature literature GO PHOSIDA GPS-Polo NetPhorest

HuPho KEGG Pfam

NetworKIN STRING phos_pred

Figure 3. Network of phosphorylation site predictor tools and the resources used to make predictions. Tools are colored purple while the databases used by the tools are colored blue.

The usability of each tool differs based on purpose of use. For possible kinases phosphorylating a single substrate of interest, web-based tools would suffice. However, the limit on the number of sequences submitted for prediction and the lack of downloadable results prevent these same tools for being useful in large-scale studies. Furthermore, downloadable tools are useful for large-scale studies, but tools can be difficult to install and use. For example, phos_pred requires modifying MATLAB code to run. NetPhos is downloadable but can only be run on Linux, while PhoScan can only be run on Windows machines. Finally, tools like GPS and phos_pred provide pre-defined cutoffs for results, while others like musite and KSP-PUEL allow users to define their own thresholds or to train the models using their own data. For large-scale kinase-substrate prediction, only 12 pre-trained tools were available that provide downloadable results. The best, unbiased way to test these tools is to use validated sites that were not used for the training of any tool. Unfortunately, most tools do not report the actual sites used for training and finding a set of sites to fit these criteria is nearly impossible. Therefore, all 12 tools were tested using gold-standard

16 positive and negative human sites from dbPTM for five kinases. The outcomes might be slightly biased in favor of newer tools and those that used some of these sites in their training. ROC curves for four kinases (CDK1, CK2, MAPK1, and PKA) are shown in Figure 4, while the fifth curve (PKC) is shown in Appendix A (Figure S1). Notably, musite was unable to predict for a few random protein sequences in each submission. MusiteDeep and GPS had the highest area under the curve (AUC) for all kinases tested. The PKA- specific tool pkaPS also performed well. Performance for most tools, however, varied across kinases.

A B 1.0 1.0 0.8 0.8 0.6 0.6 MusiteDeep:0.97 MusiteDeep:0.98 Musite:0.94 Musite:0.82 GPS:0.94 GPS:0.91 0.4 iGPS:0.61 0.4 iGPS:0.65 Netphorest:0.94 Netphorest:0.87 True positive rate positive True NetworKIN:0.92 rate positive True NetworKIN:0.83 NetPhos:0.5 NetPhos:0.86 0.2 PhosphoPredict:0.89 0.2 PhosphoPredict:0.8 PhosphoPICK:0.68 PhosphoPICK:0.79 PhoScan:0.92 PhoScan:0.86 phos_pred:0.76 phos_pred:0.64 0.0 0.0

C 0.0 0.2 0.4 0.6 0.8 1.0 D 0.0 0.2 0.4 0.6 0.8 1.0 1.0 1.0

False positive rate False positive rate 0.8 0.8 0.6 0.6 MusiteDeep:0.99 MusiteDeep:0.98 Musite:0.93 Musite:0.92 GPS:0.98 0.4 GPS:0.96 0.4 iGPS:0.63 iGPS:0.65 Netphorest:0.96 True positive rate positive True Netphorest:0.88 rate positive True NetworKIN:0.91 NetworKIN:0.84 NetPhos:0.92 0.2 PhosphoPredict:0.910.2 PhosphoPredict:0.92 PhosphoPICK:0.72 PhosphoPICK:0.77 PhoScan:0.91 PhoScan:0.91 phos_pred:0.49 pkaPS:0.97 0.0 0.0

0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0

False positive rate False positive rate

Figure 4. ROC curves for substrate prediction of four kinases. The false positive and true positive rates of substrate prediction for A) CDK1, B) CK2, C) MAPK1, and D) PKA. The AUC for each tool is listed next to the tool name. The tool pkaPS only predicts for PKA, while phos_pred does not predict for PKA and NetPhos does not predict for MAPK1.

17 Comparison of Kinase Activity Tools Using known or predicted kinases for phosphorylation sites, kinase activity can be inferred from global phosphoproteomic data. Tools and methods have been developed to predict kinase activity, but there has been little effort spent towards comparing these tools or determining the most biologically-relevant set of parameters. The available tools (PHOSIDA, KEA2, KSEA App, PHOXTRACK, and IKAP) each use a different algorithm to infer activity (Table 4). The PHOSIDA de novo motif finder uses a simple method of bootstrapping to determine enrichment of sequence motifs in a set of phosphorylated peptides and then matches those to known kinase motifs91. Kinase Enrichment Analysis 2 (KEA2) uses over-representation analysis to determine enrichment of kinase substrates in a condition64. Similarly, the KSEA App uses mean phosphorylation of substrates of kinases as a proxy for activity45. PHOXTRACK modified pre-ranked GSEA to determine enrichment of known kinase targets65. IKAP extended these methods using a cost function to infer the relative contributions of multiple kinases acting on the same site68.

Last Tool Prediction Type Method Input Type Reference Update PHOSIDA Motif 2011 sequence motifs bootstrap phosphosite 13mer Web 91 Finder KEA2 2012 kinase activity Fisher’s exact gene symbols and Web|DL|Tool 64 test phosphosite CellNOpt 2013 signaling networks logic formalisms phosphoproteomic Tool 132 data Sorad 2013 time-course ordinary phosphoproteomic Tool 133 analysis differential data equations PHOXTRACK 2014 kinase activity GSEA phosphosite 13mer Web|DL 65 and log2 expression ProteomeScout 101

CLUE 2015 time-course kinase k-means phosphoproteomic Tool 100 activity clustering data PhosFox 2015 phosphorylation comparison phosphoproteomic Tool 134 site comparison data between groups SELPHI 2015 phosphoproteomic multiple functions phosphoproteomic Web|DL 135 data analysis data DynaPho 2016 phosphoproteomic activity modules phosphoproteomic Web|DL 136 analysis for data multiple conditions IKAP 2016 kinase activity cost function phosphoproteomic Tool 68 data KinasePA 2016 kinase perturbation directional phosphoproteomic Web|Tool 137 in multiple hypothesis testing data treatments framework KSEA 2017 kinase activity Z score phosphoproteomic Web|DL|Tool 45 data

Table 4. Kinase activity prediction and phosphoproteomic dataset analysis tools. GSEA – gene set enrichment analysis

Comparison of these tools is challenging because they use different input and underlying databases. Because PHOSIDA is only available online without downloadable results, I excluded this tool from further analysis. KEA2 requires a set of sites in the format of HGNC symbol and phosphorylated amino acid residue position separated by an underscore. It contains sets for 250 different kinases. KSEA App requires a strictly

18 formatted comma-delimited file with the HGNC symbol, phosphorylated position, and non- log-transformed fold change. Users can choose between known sets from the July 2016 release of PhosphoSitePlus or the known + predicted site sets from PhosphoSitePlus and NetworKIN. PHOXTRACK requires a two-column file with a thirteenmer peptide and log- transformed fold change. It can use substrate sets from the four main databases or a user-supplied database. Finally, IKAP required tabular data entered into MATLAB, manual modification of MATLAB code to change parameters, and allowed a user to upload their own set of substrates. Because one thirteenmer might match multiple proteins and phosphorylated positions, the actual substrate list presented to each tool may differ slightly. I used a phosphoproteomic experiment with 20 kinase inhibitors to compare the kinase activity predictions by KEA2, KSEA app, PHOXTRACK, and IKAP. To determine how well each tool covered the known targets of kinases, I counted the number of significantly downregulated known kinases of each inhibitor and the significantly downregulated kinases of each inhibitor that were not known targets of that inhibitor. The KSEA App made the most true positive predictions across all experiments, while IKAP made the fewest true positive predictions (Figure 5A). PHOXTRACK made the fewest false positive predictions (Figure 5B).

A B Number of True Positives Number of True Number of False Positives Number of False 0 2 4 6 8 10 12 14 0 20 40 60 80 100 KSEA KEA2 IKAP PHOXTRACK KSEA KEA2 IKAP PHOXTRACK Figure 5. TrueKSEA and false KEA2 positiveIKAP predictions PHOXTRACK for kinase activityKSEA prediction KEA2 tools.IKAP A) PHOXTRACK For all 20 inhibitors, the number of known targets predicted to be significantly downregulated by each tool. B) For all inhibitors, the number of all significantly downregulated kinases that do not match known inhibitor targets.

Comparison of Kinase Activity Inference Methods Because the KSEA app and PHOXTRACK both maximized true positive results while minimizing false positive results, I performed a systematic comparison of the methods underlying these two tools following a protocol similar to Hernandez-Armenta et al in their benchmarking paper44. While kinase prediction tools can increase the number of substrates for each kinase, it is not known how including these substrates affects kinase activity inference. I compared the AUC for the positive set of inhibitor-target pairs and 20 different randomly permuted inhibitor-target pairs for kinase activity inference

19 using substrate sets from different levels of evidence. The GSEA and z score methods had similar performance, but the GSEA method performed better when the substrate sets came from in vitro experiments (Figure 6A, p = 0.002). For both methods, using substrate sets from in silico predictions performed significantly worse in comparison to using experimentally validated substrate sets (p = 0.0009 for GSEA and p = 0.001 for z score). As shown above, many databases collect substrates of kinases, so I compared the performance of each database paired with PhosphoSitePlus to maximize the number of kinase targets with substrate sets. The z score method performed better than the GSEA method for all databases (Figure 6B, p < 0.002). Within each method, performance across most of the databases was similar except for Signor. Substrate sets from the combination of PhosphoSitePlus and Signor performed worse against PhosphoSitePlus alone, Phospho.ELM, and Swiss-Prot using the GSEA method (p < 0.01). Substrate sets from the combination of PhosphoSitePlus and Signor performed worse against Swiss- Prot and the combination of Swiss-Prot, PhosphoSitePlus, and HPRD using the z score method (p < 0.002). Swiss-Prot performed better than PhosphoSitePlus alone and with Phospho.ELM (p < 0.04) using the z score method.

A B 1.00 1.00

0.75 0.75

0.50 0.50 AUC AUC

0.25 0.25

0.00 0.00 inin vivo vivo + - - + + PSPPSP + + + + + + + inin vitro vitro - + - + + P.ELMP.ELM - - - + - - + ini nsilico silico - - + - + HPRDHPRD - + - - - + + Swiss-ProtSwiss-Prot - - + - - + + SignorSignor - - - - + - + Method GSEA Zscore Method GSEA Zscore

Figure 6. Kinase activity AUC for the GSEA and z score methods using various substrate sets. A) AUC values for 20 randomly permutated negative sets matched with the single positive set using two methods. Substrate sets were generated using substrates with in vivo, in vitro, and/or in silico evidence. B) AUC values for experiments using substrate sets from combinations of databases.

Phosphoproteomic Experiment Analysis Tools Besides activity prediction, phosphoproteomic data can be used for other analyses. PhosFox compares phosphorylated peptides between conditions134. SELPHI allows biologists to quickly and easily analyze phosphoproteomic data with clustering analyses, kinase-substrate correlation, and pathway enrichment135. Finally, a set of tools

20 (CellNOpt, Sorad, CLUE, DynaPho, and KinasePA) were developed specifically for phosphoproteomic time-course or multiple condition analyses (Table 4)100,132,133,136,137.

Prediction of Mutation Effect Mutations can affect kinase function or presence of a phosphorylation site. PhosSNP is a database of known gene polymorphisms near phosphorylation sites that are categorized based on suspected effect138. The remaining four resources are tools to predict the effect of mutations (Table 5). Mutations Impact on Phosphorylation (MIMP) uses Bayesian statistics to predict whether mutations around a phosphorylation site will change which kinase binds to that site139. It can predict rewiring for 124 kinases using experimentally validated data, or it can be extended to predict for 322 kinases using predicted kinase-substrate relationships. ReKINect also predicts rewiring from mutations, but it further predicts the destruction or creation of phosphorylation sites and inactivation or constitutive activation of kinases140. PhosphoPICK-SNP is also similar to MIMP. It predicts the kinase responsible for phosphorylating a site, and whether a mutation affects its ability to phosphorylate the site141. Finally, wKinMut2 predicts which mutations on kinases contribute to disease142.

Last Kinases/ Method of Tool Version Prediction Type Method Reference Update Phosphatases Access PhosSNP 2009 1.0 SNVs that might influence rules Tool 138 phosphorylation status MIMP 2015 missense SNV impact on Bayesian 322 Web|DL|Tool 139 kinase-substrate model ReKINect 2015 effect of SNV on signaling PSSM Web|DL 140 network PhosphoPICK- 2016 effect of SNV on Bayesian 107 Web|DL 141 SNP phosphorylation level models wKinMut2 2016 2 SNV effect on kinase Random 450 Web|DL|Tool 142 Forest

Table 5. Resources for studying the effect of mutations on kinases and phosphorylation sites. SNV – Single Nucleotide Variation, PSSM – Position Specific Scoring Matrix

Resources for Kinase Inhibitors Which small molecules inhibit which kinases is important when considering kinases as therapeutic targets. Most resources connect known drugs to their known kinase targets (Table 6). KidFamMap connects kinases, their inhibitors, and associated diseases143. DrugKiNET shows the known inhibitors for kinases, and the kinases that a compound inhibits. It also predicts which kinases a drug can inhibit. K-Map extends these interactions to suggest the best compound to inhibit a set of kinases144. Finally, KinomeSelector groups kinases by sequence similarity and similarity of drug response. It then allows a user to choose a subset of kinases to target that cover the kinome121.

21 Last Method of Tool Description Kinases Inhibitors Reference Update Access KIDFamMap 2012 kinase-inhibitor interactions 399 35,788 Web 143 K-Map 2013 best inhibitor for a set of kinases 300 or 442 178 or 72 Web|DL 144 KinomeSelector 2014 minimal set of kinases to inhibit >500 NA Web|DL 121 DrugKiNET 2017 Known and prediction drug activity on >800 Web|DL kinases

Table 6. Kinase-inhibitor relationship resources. K-Map has two different databases – one with 178 drugs inhibiting 300 kinases and one with 72 drugs inhibiting 442 kinases.

Other Kinase Signaling Tools The final set of bioinformatics tools related to kinase signaling, summarized in Table 7, cover visualization, data retrieval, and prediction tools. KinMap, PyTMs, and RegPhos2.0 are visualization tools for the kinome tree, 3D structures of phosphorylation sites, and signaling networks, respectively95,145,146. RegPhos2.0 also provides heatmaps for kinase and substrate mRNA expression in cancer. PhosphoLogo is used to generate sequence logos for kinases147. For data retrieval, RLIMS-P and eFIP are both tools that extract data on phosphorylation interactions from the literature148,149. CPhos identifies phosphorylation sites that are conserved across species150. 14-3-3-Pred predicts phosphorylation sites in protein sequences that might bind to 14-3-3 proteins151. KinConform takes structure files and predicts whether any kinase chains in the structure are inactive or active152. Kinannote predicts whether a protein sequence is a kinase153. Finally, CrossCheck can identify the overlap between a given list of genes and the data in a database154.

Last Method of Tool Version Type Input Output Reference Update Access PhosphoLogo 2012 generate peptide logo Web|Tool 147 sequence logos sequences CPhos 2012 1.3 identifies phosphopeptides conservation Web|DL|Tool 150 conserved score phosphorylation sites RegPhos2.0 2013 2.0 visualization of gene names network Web|DL 95,96 kinase data visualization or cancer gene expression eFIP 2014 returns gene names or publications Web 149 publications words matching those involving words phosphorylation RLIMS-P 2014 2.0 returns protein PMIDs or kinase, Web|DL 148 phosphorylation keywords substrate, and information from site literature PyTMs 2015 1.2 pyMOL plugin to protein models, PTMs Tool 146 add PTMs to PTMs integrated in protein models protein models 14-3-3-Pred 2015 predicts 14-3-3 protein predicted 14-3- Web|DL 151 binding sequences 3 binding sites phosphosite KinMap 2016 kinome tree kinases tree with Web|DL 145 visualization highlighted branches KinConform 2017 determines which structures active or Tool 152 structures are inactive kinase kinases chains

22

Kinannote 2017 identifies and protein kinase Tool 153 classifies kinases sequences classification CrossCheck 2017 identifies overlap protein names protein overlap Web|DL|Tool 154 between a database and input gene list

Table 7. Miscellaneous kinase signaling tools.

Discussion The available databases and tools for studying kinase signaling cover diverse functions and include information on enzymes and their substrates, inhibitors, activity, and mutations. Together the tools comprise the current best standard for studying kinase signaling. Through the review of available resources, the human kinome and phosphatome were identified, substrate prediction tools were compared, and the choice of substrate sets on kinase activity inference was evaluated. Overall, these tools allow a researcher to discover vast amounts of information from their phosphoproteomic data and some tools can even perform entire sets of analyses with a single button click135. Despite the work that has been done, there is lots of room for advancement both in tool and method development to study kinase signaling using phosphoproteomic data. First, the majority of tools focus almost exclusively on the study of protein kinases. However, phosphatases are critical components of the kinase signaling cascade and are frequently dysregulated in cancer. Understanding the role of the interplay between kinases and phosphatases on the net phosphorylation seen in global phosphoproteomic data is essential to identifying abnormal cell signaling in disease. Furthermore, while the current tools and research are aimed at studying dysregulated protein phosphorylation, non-protein phosphorylation is also often altered in disease. For example, hexokinases, which phosphorylate glucose, drive glucose metabolism and contribute to tumor initiation in mouse models of lung and breast cancers155. The development of resources and tools to study non-protein kinases and phosphatases could advance research in a variety of fields. While the current tools provide critical functions, their error rate and accuracy could be improved. Errors are frequently propagated or amplified when tools collect data from a variety of resources. However, the integration of several databases did not affect kinase activity inference compared to a single database, so the error rate in phosphorylation site databases may have a minimal effect on downstream tools. Current tools to predict substrates of kinases perform well, but accuracy varies based on kinase. Furthermore, most tools can predict only for few kinases. MusiteDeep achieved high accuracy using their deep learning approach, but the large number of substrates required for training their method resulted in predictions for only 5 kinases. This limits its use in studying global kinase signaling. pkaPS also performed well, although it was built for a single kinase. This tool was unique because their negative set was sites experimentally validated to not be phosphorylated by PKA. Most other tools use sites not currently known to be phosphorylated by the kinase of interest, which means it is possible those sites could actually be phosphorylated by the kinase. Replicating the strategy for gold standard negative sets from pkaPS might improve accuracy for kinases in other tools.

23 Kinase activity inference is currently limited by the substrate number for kinases. The majority of kinases have fewer than 10 known substrates and the probability of identifying those sites in phosphoproteomic experiments is low. However, the current tools for adding substrates using prediction decreased the accuracy of kinase activity inference. The best two methods for activity inference are GSEA and z score. However, the substrate sets have a significant effect on the accuracy of the results. Because publicly available tools each use a different substrate set, this might affect the results when using different tools. In this review, I counted the number of true positives and false positives to compare the tools to understand how well each tool covered the target space of the inhibitors. The overrepresentation analysis method had a high number of false positives and the tools using the GSEA method and the z score method performed better. However, both could only predict downregulation for a small fraction of the known targets of those inhibitors. More work should be done to elucidate substrate sets and identify new ways to infer kinase activity. For all tools, usability can be an issue, both for bioinformaticians and biologists with no computational experience. Tools are frequently platform-dependent, do not allow downloadable results, and are not well annotated. Furthermore, tools are difficult to compare or to use more than one during analysis. The input and output formats are not standardized and use a variety of protein naming conventions. The largest challenge was deciphering input limitations and understanding results. For example, submitting a sequence with a large number of phosphorylatable residues to GPS caused the software to stall without an error message and no documentation mentioned a size limit. Musite did not provide results for a sequence or two each run without explanation. Furthermore, downloadable result files for many tools had no column headers so the column contents were unknown. For example, the downloadable file from musite has no column titles, so you have to check the table on the website to understand the results. Additionally, scores are usually presented without explanation. Only careful reading of the manuscript or the manual elucidates what value signifies a “good” response. For example, in Scansite, the score 0 is the best, with scores closest to 0 indicating the best match. But in PhosphoPICK, the score indicates probability of being phosphorylated by a kinase at that site so a score closer to 1 is better. Experts in machine learning might understand the score without explanation, but biologists likely will not. One way to fix this challenge is to have a detailed, easy-to-find manual. The manual should include ways to run the tool, the underlying mechanism of the method, and detailed description of the results. The description of the results should also be available where results are visualized. Furthermore, sample input is helpful for a new user to test the tool and determine whether the results will be useful for their experiment before preparing their own data files. In conclusion, there are many tools and resources that can be used to study kinase signaling and these tools will become even more essential with the continued production of phosphoproteomic data. It is essential for the biological community to research under- studied enzymes and to validate specific substrates of kinases and phosphatases. Furthermore, bioinformaticians should consider creating tools that utilize information from both sides of the enzymatic phosphorylation reaction. Finally, resources should be

24 carefully planned, easy to use, and well maintained. The community should work to standardize the use of enzyme IDs and phosphorylation site location.

25 CHAPTER 3

PROTEOMIC LANDSCAPE OF KINASE SIGNALING IN CANCER

Introduction The massive sequencing efforts identifying the genomic and transcriptomic alterations in numerous cancer samples by groups such as TCGA, METABRIC, and ICGC provided novel insights into cancer processes and better understanding of the underlying biological mechanisms156,157. However, the size and complexity of the data make translating these findings into the clinic challenging. Furthermore, discoveries at the genomic level do not always translate well to the protein level. The correlation between mRNA and protein expression is low for many genes158,159. Additionally, many genomic alterations are in non-coding regions and the number of druggable mutations in some cancers is low22. Because proteins are the primary drivers of cell phenotype and are the majority of drug targets in cancer, complementing the genomic data with global proteomic changes should help to clarify the effect of genomic alterations and narrow the focus to important drug targets. CPTAC undertook the goal of characterizing the proteome in three cancer types (breast, colorectal, and ovarian) that already had genomic data in the TCGA22. Additionally, they generated phosphoproteomic data for breast and ovarian cancers. To study the global proteome and phosphoproteome, they used quantitative mass spectrometry. In comparison to the antibody approach of studying proteins, mass spectrometry can discover novel peptides, elucidate the entire proteome of a sample at once, and can circumvent the non-specificity issue of antibodies. Global protein mass spectrometry is performed by first lysing tissue and digesting proteins into short peptides. Trypsin is the most commonly used digesting enzyme, but other proteases used include chymotrypsin, LysC, LysN, AspN, GluC, and ArgC160. Sequence coverage depends heavily on the selection of protease and including more than one can increase coverage160. After digestion, peptides are either labeled with a stable isotope such as isobaric tags for relative and absolute quantification (iTRAQ) or tandem mass tag (TMT) or left unlabeled. Label-free mass spectrometry is not easily quantified due to differences in peptide ionization efficiency and lack of internal standard, although counting the number of spectra correlates well with protein abundance161,162. Tags such as iTRAQ or TMT provide multiplexing and easier quantification but diminish signal and can therefore discover fewer differentially expressed proteins163,164. Peptides are further divided into fractions, usually using liquid chromatography (LC), to allow for better peptide resolution. Ideally, the mass spectrometer should detect a single peptide at a time. Tandem mass spectrometry (MS/MS), which outputs m/z and intensity values, follows sample separation. Finally, peptides are identified from the mass spectrometry spectra by matching to reference protein databases using software. Phosphoproteomics experiments are performed in much the same way, but there is an extra step after sample fractionation to enrich for phosphorylated peptides. This is usually done using immobilized metal affinity chromatography (IMAC) which attracts negatively charged phosphate groups to positive metal ions or metal oxide affinity

26 chromatography which attracts oxygen to metal atoms165. Quantification and coverage for phosphorylated peptides is lower because of lower abundance and because usually only a single peptide can be used to quantify a site compared to several peptides referring to the same protein. While mass spectrometry can identify proteins at a global level, the technique may have some bias in the quantification of proteins. Some proteins (especially smaller proteins) do not have enough tryptic peptides, might not have unique peptides for identification, or have abundance below the limit of detection. This is a particular challenge for kinases, because kinases are usually in low abundance166. Therefore, we first need to understand the limits of the technology before using it to study kinase signaling.

Methods Definition of Human Kinases, Phosphatases, and Their Substrates Kinases and phosphatases were defined as in Chapter 2. Protein lengths and sequences were downloaded from UniProt in June 2017. Experimentally validated phosphorylation sites and enzyme-substrate interactions were downloaded from dbPAF105 (http://dbpaf.biocuckoo.org/), PhosphoSitePlus (July 2017), and Signor (October 2017). dbPAF was used as a database of phosphorylation sites that did not include data from any CPTAC publications.

Cancer Proteomics Datasets Retrospective proteomic data for breast, colorectal, and ovarian cancers and phosphoproteomic data for breast and ovarian cancers generated by CPTAC were downloaded from the respective manuscripts. The breast cancer samples consisted of 105 (plus 3 replicates and 3 normal) tissue samples from patients with invasive breast adenocarcinoma158. These proteomic samples were generated by LC-MS/MS performed on samples digested with LysC and trypsin. Values were log2 transformed iTRAQ ratios of the sample compared to a common reference pool. The colorectal cancer dataset consisted of spectral counts of proteins in 95 (plus 5 replicates) tumor samples from patients with colorectal carcinoma159. These samples were previously digested with trypsin and quantified by LC-MS/MS using the label-free spectral counting method. The ovarian cancer dataset consisted of 174 (plus 32 replicates) samples from patients with ovarian serous adenocarcinoma167. These samples were digested with trypsin, processed using LC-MS/MS, and reported as log2 transformed iTRAQ ratios between the sample peptides and a common reference pool. Three unique peptides were required to identify a protein. Identified proteins used below are those reported in the respective manuscripts of the proteomic datasets. Clinical, genomic, and transcriptomic data corresponding to these samples were downloaded from cBioPortal (http://www.cbioportal.org/)168,169.

27 Phosphorylation Site Processing All quantified phosphorylation site peptides were mapped to UniProt protein sequences (July 2017). The canonical UniProt ID and sequence were chosen. If the peptide matched only to protein isoforms, one isoform ID and sequence was chosen as a representative of the group. Log ratios for multiple peptides corresponding to the same phosphorylation site were combined by median. Sites were defined as thirteenmer peptides.

Comparison with the Human Protein Atlas Expression by immunohistochemistry in multiple cancer types was downloaded from the Human Protein Atlas (HPA; www.proteinatlas.org)170 in June 2017. Categorical expression was converted into a single score for each protein in breast, colorectal, and ovarian cancers. ‘Not detected’ was zero points, ‘low’ expression was 1 point, ‘medium’ expression was 2 points, and ‘high’ expression was 3 points. Points were averaged across samples, and proteins with a score less than or equal to 0.1 were considered undetected. This corresponded to at most one sample with ‘low’ expression.

Gene Ontology Term Enrichment Gene ontology (GO) biological process term enrichment was performed using the R package version of WebGestalt. The analysis was performed for undetected enzymes in all three cancers against a background reference of the human kinome or human phosphatome. FDR with a Benjamini Hochberg correction < 0.05 was considered significant. Enrichment was also performed using Fisher’s exact test and kinase and phosphatase superfamilies, excluding non-protein kinases.

Quantified Data All analyses requiring quantified data were filtered to samples passing quality control filters as described in the original publications. Proteins and sites with missing values in fewer than 50% of the samples were retained.

Kinase Correlation with Clinical Features Kinase activity was inferred for single samples using single sample GSEA (ssGSEA) analysis in the GSVA R package for enrichment of substrate sets171,172. Comparisons of kinase activity scores, phosphorylation site levels on kinases, and kinase protein abundance between groups of patients were performed using the Wilcoxon ranked sum test. FDR < 0.05 was considered significant.

Prediction of Kinase Mutations Proteogenomic variants identified in the breast cancer manuscript were submitted to ReKINect for analysis140. Mutations were filtered to those with predicted effect on kinases and with substrates identified in PhosphoSitePlus (version May 2018). For each mutated kinase, the mean phosphorylation level of its substrates in the mutated sample was compared to samples with no mutation.

28 Statistical Analysis The R computing environment was used to perform all analyses. The package VennDiagram was used to create the Venn diagrams173. The package pheatmap was used to create the heatmaps174. Comparison of means between undetected and detected enzymes was performed using a t-test. A p value < 0.05 was considered significant.

Results Kinases and Phosphatases Identified by Mass Spectrometry Because of different experimental and data analysis methods, the total number of identified proteins differed between the proteomic datasets for breast, colorectal, and ovarian cancers (Table 8). The breast cancer dataset had the most identified proteins and consequently had the most identified kinases and phosphatases. Overall, enzymes were found across cancer samples rather than being cancer type-specific. Forty-five percent of kinases (Figure 7A) and 50% of phosphatases (Figure 7B) were identified in all three cancer types. Over 65% of the kinome and phosphatome were identified in at least two cancer types.

Cancer Type Samples Proteins Kinases Phosphatases Breast 105 15,369 622 214 Colorectal 90 7,211 347 135 Ovarian 174 9,600 435 165 Table 8. Number of identified proteins in three cancer datasets. There are 688 total known human kinases and 255 total phosphatases.

A B Colon Breast Colon Breast

32 4 5 159 4 49

309 125

1 122 2 36

3 2

56 Ovarian 33 Ovarian Figure 7. Intersection of detected A) kinases and B) phosphatases in three different cancer types. There were 56 kinases and 33 phosphatases undetected by mass spectrometry in any sample.

The other primary mechanism to measure protein expression is the use of antibodies. HPA evaluated the expression of proteins in a variety of cancer types using antibodies in immunohistochemistry. If proteins are undetected in a tissue by both mass

29 spectrometry and antibody-based methods, then the proteins are likely not expressed in that tissue. I compared the presence of proteins in breast, colorectal, and ovarian cancers in both CPTAC and HPA data. Over 1/3 of the enzymes were identified in all three cancer types by both CPTAC and HPA. However, there was little overlap between those undetected by mass spectrometry and those undetected by antibodies (Figure 8). Many enzymes that were not identified by mass spectrometry in any cancer sample had high expression by antibodies. The discrepancy could not be explained by antibody reliability. Only 17 out of 49 antibodies for enzymes undetected by mass spectrometry were annotated as uncertain, while the rest were confirmed to be specific.

A B

1. All 3 1. All 3

1. All 3 1. All 3

2. Breast and Colon 2. Breast and Colon

3. Breast and Ovarian 3. Breast and Ovarian 2. Breast and Colon 2. Breast and Colon 4. Colon and Ovarian 3. Breast and Ovarian 4. Colon and Ovarian 3. Breast and Ovarian 4. Colon and Ovarian 4. Colon and Ovarian 5. Breast only 5. Breast only 5. Breast only 6. Colon only 5. Breast only 6. Colon only 7. Ovarian only 7. Ovarian only 6. Colon only 8. Undetected 8. Undetected 6. Colon only 7. Ovarian only 7. Ovarian only 8. Undetected 8. Undetected 9. Not in assay 9. Not in assay

CPTAC HPA CPTAC HPA Figure 8. Enzymes in CPTAC and HPA data. A) Identification of kinases and B) phosphatases by mass spectrometry (CPTAC) and immunohistochemistry (HPA). A large proportion of enzymes were identified in all three cancer types in both CPTAC and HPA. Some proteins undetected by mass spectrometry (indicated in red) showed expression with antibodies in cancer tissue.

Protein Length Protein length can also affect detection by mass spectrometry. Longer amino acid sequences might be digested into more peptides, which increases the chance of detection. There was a significant difference in length of undetected compared to detected proteins in breast cancer, but not colorectal or ovarian cancers (Figure 9).

30 A Breast Kinases B Colon Kinases C Ovarian Kinases

● ● ● 8000 6000

● ● ● ● ● ● 4000 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

2000 ● ● ● Number of Amino Acids 0

Undetected Detected Undetected Detected Undetected Detected T test T test T test p value = 3.12e−11 p value = 0.123 p value = 0.819 D Breast Phosphatases E Colon Phosphatases F Ovarian Phosphatases ● ● ●

2500 ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

1500 ● ● ● ● ● ● ●

● 500 Number of Amino Acids

Undetected Detected Undetected Detected Undetected Detected T test T test T test p value = 1.66e−06 p value = 0.686 p value = 0.74

Figure 9. Number of amino acids in proteins that were and were not detected by mass spectrometry. There was a significant difference in the lengths of the A) kinases and D) phosphatases detected by mass spectrometry in breast cancer samples compared to those that were not detected. There was no difference in detected B) kinase and E) phosphatase length in colorectal cancer, nor in the C) kinases and F) phosphatases of ovarian cancer. mRNA Expression Proteins in low abundance may be difficult to identify in mass spectrometry. As a surrogate for protein expression, mRNA expression can be used to assess whether proteins have a low expression. In these three cancers, proteins that were not identified in individual cancers had a lower mRNA expression than those that were identified in that cancer (Figure 10).

31 A Breast Kinases B Colon Kinases C Ovarian Kinases

15 ● ● ● 10

● ● 5 ● ● ● ●

Log2 Expression ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● 0 ● ●

Undetected Detected Undetected Detected Undetected Detected T test T test T test p value = 1.18e−07 p value = 3.16e−33 p value = 5.46e−27 D Breast Phosphatases E Colon Phosphatases F Ovarian Phosphatases

15 ● ● 10

● 5 ● ● ●

Log2 Expression ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● 0

Undetected Detected Undetected Detected Undetected Detected T test T test T test p value = 3.14e−08 p value = 1.59e−17 p value = 1.26e−12 Figure 10. Log2 mRNA expression of proteins that were and were not detected by mass spectrometry. Difference in expression of A,B,C) kinases and D,E,F) phosphatases detected by mass spectrometry in breast cancer (A,D), colorectal cancer (B,E), and ovarian cancer (C,F).

Term Enrichment of Undetected Enzymes There was no significant enrichment (FDR < 0.05) of GO biological process terms for undetected kinases or phosphatases compared to the full human kinome or phosphatome. There was also no significant enrichment of individual kinase families in the undetected kinases. However, the undetected phosphatases were enriched for the DSP, phosphatidic acid phosphatase, and the chloroperoxidase superfamilies. Fewer than half of the members of each of these superfamilies were identified by mass spectrometry in these CPTAC experiments.

Phosphoproteomic Analysis of Breast and Ovarian Cancers The CPTAC studies for both breast and ovarian cancers contained phosphoproteomic data complementary to the proteomic data. The breast cancer data

32 identified almost 23,000 phosphorylation sites that had non-missing values in at least 50% of the 77 samples passing quality-control filters. This corresponded to sites on almost 6,000 unique proteins. The ovarian cancer data, however, only identified 5,000 phosphorylation sites on about 2,500 proteins in at least 50% of 66 samples. Most of these phosphorylation sites (92% in breast, 96% in ovarian) had been identified in previous low or high-throughput experiments as reported in the dbPAF database. These data captured phosphorylation sites on kinases (357 kinases in breast, 177 in ovarian) and phosphatases (96 phosphatases in breast, 41 in ovarian). Almost all the kinases and phosphatases with quantified phosphorylation sites had corresponding proteomic data.

Kinase Activity in Breast Cancer Kinase activity can be inferred by determining the enrichment of its substrates in a sample. Out of the almost 23,000 quantified phosphorylation sites in breast cancer, only 1,486 have a known kinase or phosphatase. Using ssGSEA of these sites, activity for 158 kinases and phosphatases could be determined. The enzymes with the highest variability in activity (standard deviation ≥ 0.25) across samples were STK11, TTBK1, and CDK5. Kinase activity clustered with transcriptomic subtypes, suggesting similar kinase signaling in tumors with similar gene expression (Figure 11). There was a subset of primarily luminal A tumors that had higher relative activity of most kinases compared to the other samples. There was another subset of mixed luminal A and luminal B tumors that had very low relative activity of MAP2K and MAP3K kinases compared to other samples.

33 0.6 PAM50

ABL1 AKT Basal AKT1 AKT2 AKT3 AMPK 0.4 Her2 AURKA AURKB AXL BRAF Luminal A Calcineurin CAMK2A CAMK2B CAMK2D 0.2 Luminal B CDK5 CHEK1 DAPK1 DYRK2 EEF2K EGFR ERK1/2 FYN 0 GRK6 GSK3A GSK3B IKBKB IKBKE JAK2 LCK MAP3K7 MAP3K8 −0.2 MAPK1 MAPK11 MAPK14 MAPK3 MAPK8 MAPK9 MAPKAPK2 MARK2 −0.4 MERTK NLK PAK1 PAK2 PIM1 PPP2CA PPP2CB PRKAA1 −0.6 PRKAA2 PRKACA PRKCA PRKCB PRKCD PRKCE PRKCG PRKCH PRKCI PRKCQ PRKCZ PRKD1 PRKD2 PRKG1 PRKG2 PTPN1 PTPN2 ROCK1 RPS6K RPS6KA1 RPS6KA3 RPS6KB1 RPS6KB2 SGK1 SGK3 SIK2 SRC TBK1 TRPM7 ULK1 ARAF CAMK1 LATS1 LATS2 MAP2K1 MAP2K2 MAP2K3 MAP2K4 MAP2K6 MAP2K7 MAP3K11 MAP3K5 MAP3K6 MAPK7 MEK1/2 PAK4 PDPK1 PIM2 PIM3 PRKACB PRKACG RAF1 RET RPS6KA5 STK11 STK3 STK38 VRK1 CDK5/CDK5R1 CSNK1D DYRK1A EIF2AK2 MAPK12 MAPK13 MARK1 NUAK1 PKN1 PP2B TTBK1 ATM ATR CDC14B CDC7 CDK1 CDK12 CDK2 CDK3 CDK4 CDK6 CDK7 CDK9 CHEK2 CHUK CSNK1A1 CSNK1E CSNK2A1 CSNK2A2 CSNK2B CyclinA2/CDK2 CyclinB/CDK1 CyclinE/CDK2 HIPK2 IRAK4 JNK MAPK10 MELK MTOR mTORC1 NEK6 p38 PKM PLK1 PLK2 PLK3 PPP1CA PPP1CB PRKDC ROBO SIK1 TTK UHMK1 A2 AR AO A8 C8 A2 AR A2 A2 A2 BH E2 BH A2 A8 AN D8 A2 E2 E2 AR C8 AO AN AN A2 A2 A2 A7 A8 C8 C8 AO C8 C8 BH AO BH AO AR AR AO BH BH A8 C8 BH AR AO AN A8 AR BH AO AO C8 A2 AO BH BH A2 A2 AN A7 AO A2 AO AO A2 AR A8 C8 A7 AN BH AR C8 − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − A0YC A08Z A0EV A0YD A0T6 A0EX A154 A0EY A076 A0D2 A158 A15A A0EQ A0YG A0CM A0CE A09G A079 A06Z A0SW A0YM A0SX A13F A0YF A0T3 A06N A0CJ A135 A142 A12V A134 A12L A131 A12Z A138 A130 A12T A12U A1AS A0TX A0E9 A18N A0AM A0U4 A0FL A0AL A18Q A0E1 A0TT A1AP A0AV A18U A0DG A0TV A04A A1AV A0C1 A0BV A0C7 A0FK A0TR A0AJ A0DD A1AW A0JC A12F A12D A0JE A0JJ A0J6 A03O A0JM A12B A0JL A0J9 A126 A12E

Figure 11. Heatmap of inferred kinase and phosphatase activity scores for breast cancer samples. Kinase activity was determined using ssGSEA of enzyme site-level substrates. Red indicates increased kinase activity relative to other samples, while blue indicates decreased kinase activity. The sample PAM50 subtype is indicated by black for basal, green for Her2, red for luminal A, and blue for luminal B. Rows are ordered by k means cluster (4 clusters).

34

To determine the effect of kinase activity on patients, I wanted to compare activity to clinical features, but unfortunately clinical variables were limited in this cohort. Therefore, I chose two interesting data features: early vs late stage tumors and tumors with and without GATA3 mutation. The transcription factor GATA3 is mutated in about 10% of breast tumors, but the phenotypic effects are not well understood175. None of the kinases with highly variable activity were significantly different in any of these groups. However, the activity of MERTK was significantly lower in samples with GATA3 mutation compared to samples without GATA3 mutation (Figure 12A, FDR = 0.02). No kinase activity scores were significantly different between early (stages 1 and 2) tumors and late (stages 3 and 4) tumors. However, the protein level of the kinase NT5C was significantly increased in stage 1&2 compared to 3&4 (Figure 12B, FDR = 0.01). NT5C is a kinase and therefore does not have an activity score. No phosphorylation levels on any kinase were significantly different among these groups.

A

GATA3 4 GATA3 4 WT GATA3CNA GATA3 Mutation CNA 2 WT 2 NA RNA Mutation RNA 0 CNANA Protein None -−22CNA Activity 0 Deletion None -−44 Amplification Protein B −2

−4 66 Stage StageStage 6StageStage CNAStageActivity Stage1 CNACNA 44 4 Stage2Stage1Stage1 RNA CNA 2 Stage2Stage2 RNARNA 2 2 None Deletion 00 CNAAmplification ProteinProteinProtein 0 0 CNA -−22 NoneNone p−pS182−S182 −2−2 DeletionDeletion p−S182 -−4−44 p−pT99−T99 −4 AmplificationAmplification −6 -6−6−6 Figure 12. Multi-omics profile of MERTK and NT5C in breast cancer. A) pRN−T99A, protein, and ssGSEA activity scores for MERTK. B) RNA, protein, and phosphorylation levels at S182 and T99 for NT5C. All values were standardized using z score.

Effect of Mutations on Kinase Signaling Finally, I wanted to determine the effect of protein mutations on kinase activity in breast cancer. To do this, I predicted the effect of the 3,658 identified protein variants using ReKINect. Of the 641 mutations that had a possible effect on kinase signaling, most were predicted to have an unknown function (Table 9). Phosphoproteomic data of kinase substrates could be used to evaluate the effect of these mutations. Unfortunately, mutations are unique to individual patients and proteins, which limits the statistical

35 analysis. However, some mutations show interesting patterns worth exploring in future work. For example, sample A2-A0YD had a D110N mutation in the kinase domain of CDK6. ReKINect predicted this mutation to be an “uninterpreted mutation on kinase domain.” The mean phosphorylation levels of CDK6 substrates for that sample were 1.8 standard deviations below the mean for all samples. Therefore, this mutation might inhibit the function of CDK6. Similarly, patient A0-A126 had a T337K mutation on MASTL. Phosphorylation of its substrate was 1.5 standard deviations below the mean.

Mutation Type Frequency Destruction of phosphorylation site 64 Kinase downstream rewiring on kinase domain 1 Kinase downstream rewiring, Mutation around phosphorylation site on kinase domain 1 Kinase inactivation by phosphorylation destruction, Destruction of phosphorylation 1 site on kinase domain Mutation around phosphorylation site 510 Phosphomimicking mutation 9 SH2 downstream rewiring on kinase protein on SH2 domain 1 SH2 downstream rewiring on SH2 domain 1 Uninterpreted mutation on kinase 41 Uninterpreted mutation on SH2 12 Table 9. Predicted proteogenomic mutation effect from ReKINect.

Discussion The proteomic and phosphoproteomic data generated by CPTAC provide unique opportunities to explore signaling dysregulation in cancer. Although mass spectrometry experimental protocol and analysis methods have a significant effect on the number of protein identifications, in general it can identify all aspects of kinase signaling with limited biases. Notably a minimal number of kinases and phosphatases were not detected in any of the three experiments, but there was no single reason for exclusion. Some of these enzymes are likely to be tissue specific (e.g., GRK7 is retina-specific) or exhibit low expression in these cancer tissues. Because there was no bias in identifiable kinase family, we can be confident proteomic data can be used to study global kinase signaling in cancer tissue. Phosphoproteomic and proteomic data can identify individual kinases that are interesting in breast cancer. While these are primarily association studies and limited by sample size, they provide a starting point for future studies. In particular, STK11, CDK5, and TTBK1 appear to have differential activity across samples and may become targets of individualized treatment. STK11 encodes liver kinase B1 (LKB1), which is a tumor suppressor involved in major cell pathways. In breast cancer, increased LKB1 activity can inhibit TGF-b1 transcription, leading to inhibition of epithelial-to-mesenchymal transition (EMT)176. Furthermore, lower LKB1 mRNA expression correlated with poor survival in the TCGA cohort177, although another study found significant correlation with survival only in HER2-positive patients178. Few reports explore CDK5 or TTBK1 in breast cancer, but one study suggested lower CDK5 expression promotes longer metastasis-free survival179 and CDK5 is pro-tumorigenic in several cancer types180.

36 In addition to the kinases with variable activity across samples, predicted MERTK activity was significantly reduced in samples with mutations in GATA3. Although no associations between these two proteins are noted in the literature, it is possible that GATA3 affects MERTK expression. MERTK is predicted to be a target of GATA3 based on its motif181 and MERTK mRNA was significantly reduced in GATA3 knock- out keratinocytes182. Unfortunately, clinical outcomes are limited in the breast cancer cohort, but an interesting finding was the correlation of NT5C abundance with breast cancer stage. Little is known about NT5C in breast cancer, but in leukemia patients treated with , higher mRNA expression of NT5C correlated with worse outcomes. This finding is counter to its high expression in early stage breast cancer here183. Finally, the use of phosphoproteomic data to validate kinase signaling mutations holds promise. Future work to quantify both variant and wild-type peptides will help determine the effect of mutations near a phosphorylation site on the ability of that site to be phosphorylated. Furthermore, extending mutation analyses to include SNVs identified at the gene level might increase sample size and allow for statistical analysis of the effect of individual kinase mutations on the activity of that kinase.

37 CHAPTER 4

CHARACTERIZATION OF MOLECULAR SUBTYPE-SPECIFIC DRIVER SUBNETWORKS IN BREAST CANCER

Introduction Breast cancer is a heterogeneous disease with different treatment modalities and outcomes. To differentiate between patients, breast cancer is split into various subtypes by histological type, stage, presence of hormone receptors, or mRNA expression. Using mRNA data, breast tumors can be classified into four different molecular subtypes based on the expression of 50 genes184,185. The four PAM50 subtypes, basal, Her2, luminal A, and luminal B, have different characteristics, survival rates, and treatment options186,187. While the expression patterns of tumors within a subtype are similar, the underlying signaling mechanisms driving these subtypes are not well understood. Cancer is hypothesized to be driven by genomic mutations that lead to downstream signaling dysregulation and phenotype changes. However, most mutations are present in small percentages of patients. As seen in Chapter 3, mutations are frequently specific for individual patients. However, individual mutations in patients might all converge on the same pathway. For example, one patient might have an inactivating mutation on PTEN, which results in active AKT signaling. A second patient might have an activating mutation on PI3K, which also results in active AKT signaling. In this way, different individual mutations could drive the same altered expression patterns within a subtype. One way to extract signaling networks from individual mutations is to use the random walk with restart (RWR). In this method, a random walk on a PPI network starts from mutated nodes and randomly walks to neighboring nodes at each time step. The restart value allows the random walk to teleport back to the starting node to restrict the final smoothed network to the nearest neighborhood of the starting nodes. The final probabilities of landing on individual nodes provides a smoothed network from the starting mutated nodes. Because the probability of landing on a single node immediately downstream of two different mutated genes is high, the most significant genes will be those downstream of multiple different mutations. This method has been used to identify subtypes of various cancers, extract driver subnetworks in colorectal cancer, and identify altered signaling pathways after drug treatment, among others188–190. The limitations to identifying driver subnetworks using RWR are validation and identifying the network direction. Mutations can be activating or inhibitory on a gene and the actual function of most individual mutations in cancer is not known. Furthermore, validating the network as a true driver of the subtype usually requires extensive molecular biology experiments. Phosphoproteomic data, however, provides a unique opportunity to both validate the extracted subnetworks and determine whether the network is under- or overactive in that subtype.

38 Methods Protein-Protein Interaction Network The PPI network was downloaded from High-quality INTeractomes191 (HINT, http://hint.yulab.org/) in May 2017. The largest connected component was extracted using igraph192 in R and consisted of 11,711 nodes and 110,838 edges. The nodes were normalized by degree.

Starting Probabilities Non-silent mutation data were downloaded for 806 breast cancer samples from cBioPortal and processed as in Chapter 3. Mutation data for individual samples were normalized by total number of mutations within a sample, summed across all samples within a PAM50 subtype, and averaged. Nodes with higher probability indicated more samples with that mutation. Genes were randomly permutated 1000 times for statistical analysis.

Driver Subnetwork Extraction Starting probabilities for all mutated nodes in a subtype and the permuted probabilities were submitted to the NetWalker190 algorithm implemented in R. The restart probability was set to 0.5. Local p values were calculated as the rank of the gene in the real run across all permuted runs. Global p values were calculated as the rank of the gene across all scores. Nodes were considered significant with both a global and local p < 0.05. Networks were generated by filtering for edges in the HINT network between significant nodes. Additional subnetworks were extracted using edges between significant nodes that were unique to a subtype.

Driver Subnetwork Pathway Enrichment Overrepresentation enrichment analysis of significant nodes in each network was performed using WebGestaltR and WikiPathways193. The minimum overlap was set to be 10 and the maximum was 500. The background was all possible nodes in the HINT network and pathways were considered significant with FDR < 0.05.

Differential Phosphorylation Site Abundance Phosphorylation sites were processed as in Chapter 3. Log2 ratios were converted to z score. The z scores for each subtype were compared to the samples of all other subtypes using the Wilcoxon signed rank test. P values were adjusted using the Benjamini-Hochberg method and significance was assigned as FDR < 0.05.

Evaluation of the DNA Damage Response in the Basal Subnetwork The number of nodes with significant differential phosphorylation were counted for each significant pathway. The pathways with the highest number of these nodes were explored. The DNA damage response pathway (WP707) in the basal subnetwork had the highest number of significant differentially phosphorylated nodes. Pathway direction was inferred from the regulatory effect of phosphorylation on those nodes.

39 Differential Substrate Phosphorylation The kinases, defined as in Chapter 2, were extracted from each subnetwork. For every kinase in an individual subnetwork with at least 1 substrate in the phosphoproteomic data, the z score phosphorylation levels of the substrates were compared between samples in a single subtype and all other samples using a Wilcoxon rank sum test. P values were adjusted using the Benjamini-Hochberg method and were considered significant at FDR < 0.05.

Software Packages The R computing environment was used to perform all analyses. The Venn diagram was created using Venny v2.1. Cytoscape194 was used for network visualization and the R package pheatmap was used to create the heatmap.

Results Generation of Driver Subnetworks Subtype-specific driver subnetworks were created using network propagation from mutated genes in that subtype. There were 145 basal samples, 66 Her2, 416 luminal A, and 179 luminal B. The mutations in each sample were normalized by number of mutations in the sample and were combined within a subtype. For all subtypes, the network propagation started from about 4000 mutated nodes. Approximately 600 nodes in each subtype had statistically significant steady state probability after propagation. Over 50% of the nodes for each subtype were unique to that subtype (Figure 13). Only 29 genes were significant in all four networks and only 6.5% of the genes were present in at least 3 of the 4 networks, indicating subtypes had uniquely altered pathways.

Her2 Luminal A

Basal Luminal B

Figure 13. Venn diagram of significant nodes in subtype-specific driver subnetworks.

I filtered for edges between all significant nodes in a subtype and extracted the largest connected component for each network. The basal subnetwork had 287 nodes and 468 edges, the Her2 subnetwork had 239 nodes and 407 edges, luminal A had 321

40 nodes and 509 edges, and luminal B had 299 nodes and 458 edges. All networks can be found in Appendix C. Using WikiPathways, I performed pathway enrichment for each of the driver subnetworks. The basal subnetwork was enriched for alpha 6 beta 4 signaling, DNA damage response, FasL pathway and stress induction of HSP, and striated muscle contraction (Table D1, FDR < 0.05). The Her2 subnetwork was enriched for signaling of hepatocyte growth factor receptor, rac1/pak1/p38/MMP-2 pathway, PDGF pathway, and focal adhesion (Table D2, FDR < 0.05). Both the luminal A and luminal B subnetworks had a large number of enriched pathways. In addition to several of the pathways enriched in basal and Her2, luminal A also had enrichment of ErbB signaling, estrogen signaling, MAPK, and PI3K-AKT-mTOR signaling pathways (Table D3, FDR < 0.05). The luminal B subnetwork uniquely had enrichment for regulation of actin cytoskeleton, miRNA targets in ECM and membrane receptors, and TGF-beta signaling (Table D4, FDR < 0.05).

Differential Abundance of Phosphorylation Sites Across Subtypes To use phosphoproteomic data to validate driver subnetworks, phosphorylation patterns should differ between subtypes. However, few sites were significantly changed in one subtype compared to the others. There were no significant sites in Her2 samples compared to the others (Figure 14B). In basal, 1338 sites were significant compared to the others, 820 were in luminal A, and 20 were in luminal B (Figure 14A,C,D). Only 71 of the significant sites were on proteins significant in the basal subnetwork, 41 in the luminal A, and 2 in the luminal B. Despite this minimal overlap, phosphoproteomic data might still suggest a direction for the pathways.

41 A B 5 5 ●

4 ● 4

● ● ●

log10(FDR)) ● log10(FDR)) ● ● ● − − 3 3

● ● ● ● ● ● ● ● ●● ● ● ●●● ● ● ● ● ● ●● ●● ● ● ●● ● ● ●● ●● ● ● ● ● ● ● ● ● ● ●● 2 ● ● 2 ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ●●● ●●●●● ●● ● ● ● ●● ● ●● ●● ●●●● ●● ● ● ● ● ● ● ● ● ●● ●● ●● ● ● ●● ●●●●●● ● ●● ● ● ●●●● ●●● ● ● ● ●●●● ●●●● ● ● ●●● ● ●● ●●● ● ●● ●●● ● ●●●● ●● ● ● ●●●● ●● ●● ●●●●● ●● ● ● ● ● ● ● ●●●● ●●●●●● ● ● ●● ●●●●●●● ● ●● ● ● ● ● 1 ● ●●● ●● ●●●●● ● ● ● ● 1 ● ● ● ● ● ● ●● ●●●●●●●●●●●●●●● ● ●● ● ●● ●●● ●●●●●●●●●● ● ●● ● ● ● ● ●●●●●●●●●● ● ●●●●●●●●●●● ● ● ●● ●●● ●●●●●● ● ● ●●●●●●●●●●●●●●●●●●●●●●● ● ● ● ● ●● ● ●●●●●●● ●● ●●●●●●●●●●●●●●●● ●● ● ● ● ● ● ●●●●●●●●● ● ●●●●●●●●●●●●●●●●●●●●● ● ● ● ● ● ● ●●●●● ●●●●●●●●●●● ●●●●●●●●●●●●●●●●●●● ● ● ●● ● ● ●●● ●●●●●●●●●●● ● ● ● ● ●●●●●●●●●●●●●● ● ● ● ● ● ● ●● ● ● ● ●●● ●●●●●●●●●●●●●●● ●●●●●●●●●●●●●●●●●●● ●●● ● ● ● ● ● ● ● ● ● ● ●●●●●●●●●●●●●●●●● ● ●●●●●●●●●●●●●●●●●●●● ● ● ●● ● ● ● ●●● ●●●● ●● ● ●●●●●●●●●●●●●●●●●●●●●● ●●●●●●●●●●●●●●●●●●●●●● ●● ●● ● ● ●●●●●● ● ● ● ●●●●● ●●●●● ● ●●●●●● ●●●●●●●●●●●●●●● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●● ● ● ● ●●● ●●●●●●●●●●●●● ●●●●●●●●●●●●●●●●● ● ● ● ● ● ●●● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●● ● ● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●● ●● ●●●●●●●●●●●●●●●●●●●●●● ● ● ● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●● ● ● ●● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●● ● ●●●●●●●●●●●●●●●●●●●●●●●●●●● ● ●●● ● ● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●● ● ● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●● ● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●● ● ● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●● ● ● ● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●● ● 0 0 Signed Rank Test ( Signed Rank Test ( Signed Rank Test

−2 −1 0 1 2 −2 −1 0 1 2 Median Difference Median Difference C D 5 5 4 4 log10(FDR)) log10(FDR)) − − 3 3

● ● ● ● ●● ● ● ● ● ● ● ● ● ● ●●

2 ● ● 2 ● ● ● ●●● ● ● ● ●● ● ●● ● ● ● ● ●● ●● ● ● ● ●● ● ● ●● ●● ● ● ●● ● ● ●● ● ● ● ● ●● ● ●● ● ● ●●● ● ● ● ● ●● ●● ● ● ●● ● ● ●● ● ● ● ● ●●●●●● ●● ● ● ●●●● ●● ● ●● ●●●● ● ● ● ● ● ● ●●●● ● ● ●●●●● ● ●● ● ● ●● ●●●●●●● ●●●● ● ● ●●●●●● ●●●● ● 1 ●●●●●● ●● ●●● ●●●●●● 1 ●● ●● ● ● ● ● ● ● ● ●●●●●●●● ●●● ●●●●●●● ●● ● ● ● ●●●●●●●●● ●●●● ●● ●● ● ●●● ● ● ●● ●● ● ●●●●●● ●●● ● ●● ● ●●●●●●●●●● ● ●●●●●●●●●● ●●●●● ●●● ●●● ●●●●● ● ● ● ●●● ●● ● ●● ●●●●●●● ●● ●●●●●●●●●●●●●●● ●●●●●●●●●●●●● ● ● ●●●●●● ● ● ● ●● ●●● ●●●●● ●● ● ●●●●●●●●●●●●●● ●● ● ●●●●●●●● ●●● ● ●●●●● ● ● ●●● ●●●●●●●●●●●●● ●●●●●●●●●●●●●●●● ●●●●● ●●●●●●●●●●● ●● ●●●●●●●●● ● ●●●●●●●●●●●●●●●●●●● ● ●●●●●●●●●●●●●●●●●●●●● ● ● ●●●●● ●●●●●●●●●●●●●● ●●● ●●●●●●●●● ●●● ●● ● ●●●●●●●●●●●●● ●● ● ●●●●●●●●●●●●●●● ●●● ● ●●●●●●●●●●●●●●●●●● ● ●●●●●●●●●●● ● ● ●●●●●●●●●●●●●●●●●●●●● ● ●●●●●●●●●●●●● ● ● ●●●●●●●●●●●●●●●● ● ● ●●●●●●●●●●●●●●●●●● ● ●●●●●●●●●●●●●●●●●●●●●●●●● ●●●●●●●●●●●●●●●●●●●●● ●●●●●● ● ● ●●●●●●●●●●●●●●●●●●●●●● ●●●●●●●●●●●●●●●●●●●●●●● ● ●●●●●●●●●●●●●●●●●●●●●●● ● ●●●●●●●●●●●●●●●●●●●●● ● ● ●●●●●●●●●●●●●●●●●●●●●●●●●●● ●● ● ●●●●●●●●●●●●●●●●● ● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●● ●● ● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●● ●●●●●●●●●●●●●●●●●●●●●●●●●● ● ●●● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●● ● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●● ●●●●●●●●●●●●●●●●●●●●● ●● ● ● ● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●● ● ● ● ●●● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●● ● ● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●● ● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●● ● ● ● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●● ●●●● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●● ● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●● 0 0 Signed Rank Test ( Signed Rank Test ( Signed Rank Test

−2 −1 0 1 2 −2 −1 0 1 2 Median Difference Median Difference

Figure 14. Relative differential phosphorylation abundance in breast cancer subtypes. Normalized phosphorylation abundance at each site on nodes in the network was compared between each subtype and the remaining samples. Red indicates significantly increased phosphorylation sites and blue indicates significantly decreased phosphorylation sites in A) basal, C) luminal A, and D) luminal B subtypes relative to all other samples. There was no significant differential phosphorylation in B) Her2 samples.

DNA Damage Response in Basal One interesting enriched pathway in the basal subtype was the DNA damage response pathway. Eleven of the nodes in the basal subnetwork were present in the DNA damage response pathway. Four of these nodes had phosphorylation sites that were significantly increased in the basal subtype compared to all other samples (Figure 3). Phosphorylation on TP53 at S315 increases in response to DNA damage and promotes TP53 activity in the response to DNA damage195. Phosphorylation of both kinases PRKDC and CHEK2 increase their activity and also promote their response to DNA damage196,197. Therefore, these phosphorylation sites indicate that the DNA damage response is highly active in basal tumors.

42

ificantly increased phosphorylation. increased ificantly . Overlap of the basal driver subnetwork and the DNA damage response pathway. response damage DNA the and subnetwork driver basal the of Overlap .

15 Figure Purple nodes indicate genes in the basal driver subnetwork and red nodes indicate genes that sign had also and subnetwork driver basal the in were

43 Subnetworks Unique for Each Subtype To determine what makes each subtype unique, I further filtered the driver subnetworks to include only edges between nodes that were unique to an individual subtype and overlapped significant changes in phosphorylation. I kept only components consisting of more than 3 nodes. The basal subnetwork consisted of 6 components with 63 nodes and 58 edges (Figure 16). Most components had at least one node with significantly altered phosphorylation in basal samples compared to all other samples. The largest component surrounded the protein estrogen receptor beta (ERb) encoded by the ESR2 gene.

ZEB2 DROSHA CHD2 TMEM57 ARHGAP9 THRAP3

TULP3 MTMR11 SMAD9 AFF1 BCLAF1 MYH7B

MYH7 ASH2L

MYOM2 RFC1 SPTBN1 PLEC SREBF1 ESR2 SHROOM3

H4 CREBBP PLEKHA5 MYH4 LAMB1 H3.2 RRP12 SNRPN RBFOX1 CRX RPL13A

ING1

MT-ND1 ALG13

AREL1 NCF2 MRPS31 RPL6 DLGAP4 DOCK9 ZFPM2 GATA4 ASXL1 UBC CDC42 RPL3L

RABGAP1 CYLD PREX1 BMX STAT3 PLAA FMNL1

IK

CASP6 CHD3 SDCCAG3 CSNK1E ARHGEF1 FAM117B BATF3 HLF

Figure 16. Nodes unique to the basal subnetwork. Nodes with significantly increased phosphorylation sites are colored red, while decreased phosphorylation is indicated by blue.

The luminal A subtype was the only other network to contain nodes with altered phosphorylation levels. It contained 6 components with 43 nodes and 39 edges (Figure 17). Unlike the basal subnetwork, most of the phosphorylation sites were downregulated. Furthermore, it contained a network for ESR1, which encodes ERa, instead of ESR2.

44 CHD4 SOS2 BCR

EPSTI1

BLID PIK3CG HDAC1 GANAB DDX5 BMPR2 HRAS

ESR1 KIF26A MAPK8 GLCE PSMC3IP IL24 DEAF1 ARAF MAPK10 HSPA8

MAPK8IP3 KIF1A

KRAS

PPARG RIC8A

THRSP TNFSF13

MSX2 BTK

ERCC6L ARHGEF4 GTF2I

LMO4 GCC1

ZFP36L2 HLA-B SNX5 VPS52 NINL

WTAP CACNA1D DCTN5

Figure 17. Nodes unique to the luminal A subnetwork. Nodes with significantly increased phosphorylation sites are colored red, while decreased phosphorylation is indicated by blue.

Altered Kinase Substrates The networks for both basal and luminal A contained a few kinases. To assess the differences in kinase signaling in these networks, I compared the phosphorylation levels of substrates of kinases in each subtype to the other subtypes. Approximately 50 kinases were present in each subtype driver subnetwork and slightly over half had substrates in the phosphoproteomic data. For each subtype, a fraction of the kinases had altered phosphorylation of its substrates in that subtype compared to the others (Figure 18). Interestingly, some of the kinases identified in the unique basal and luminal A subnetworks had substrate data. A single kinase, CSNK1E, was unique to the basal subtype. Phosphorylation of CSNK1E substrates was higher in the basal samples compared to others. The unique luminal A subnetworks had 8 kinases, 5 of which had phosphorylation data. The substrates of ARAF were significantly increased, while the substrates of MAPK10 and BCR were significantly decreased. The best validation of the identified subnetworks would be if the kinases in the subnetworks were dysregulated only in those subtypes. This was not the case, as the substrates of about 10 kinases not identified in each driver subnetwork were also significantly dysregulated in that subtype.

45 CDK12 WNK1 MAP3K1 10 FYN CHEK2 TAOK1 TNIK PRKG2 5 LMTK2 ABL1 WNK2 CDK19 ALK 0 PTK2B BMX PASK PRKD1 −5 PIK3CD MAP3K12 RPS6KA3 CSNK1E MERTK ATM LIMK2 SGK1 PKMYT1 PRKDC MAPK14 MTOR ERBB2 BCR MAPK10 SMG1 CHUK PRKD2 BTK MAP4K1 MAPK8 TAF1 PRKCB JAK1 CIT ARAF MAP3K5 IKBKB MAP2K4 RET MAP2K3 AXL ROCK1 MYLK PRKCE DAPK1 SIK3 DYRK3 IGF1R MAPK4 PINK1 MAP4K5 RPS6KA2 MAPKAPK3 PKN1 DYRK1A PDGFRB DMPK GRK2 FGR RIPK3 CDK17 PRKACG ABL2 EGFR JAK2 SRC PAK3 FGFR4 PAK1 PLK2 PLK1 MAP3K10

Basal Her2 Luminal A Luminal B basal Her2 luminal A luminal B Figure 18. Relative phosphorylation of substrates of kinases. Red indicates significantly increased phosphorylation, blue indicates significantly decreased phosphorylation, and white indicates no change in phosphorylation. Color is based on -log10(FDR). Gray indicates kinases that were not significantly altered in the driver subnetworks. Grey indicates the kinase is not present in that driver subnetwork.

46 Discussion Overall, genomic mutations converged on distinct subnetworks in breast cancer molecular subtypes. While phosphoproteomic data had limited utility in validation due to minimal phosphorylation differences and low signal, the data provided interesting insights into the function of these networks. The driver subnetworks extracted for each subtype covered known dysregulated pathways. For example, the estrogen receptor signaling pathway was enriched in the luminal A subnetwork. Luminal A samples are typically ERa-positive and respond well to endocrine therapy198. Furthermore, PI3K is most frequently mutated in ER-positive luminal tumors and this pathway was also enriched in the luminal A subnetwork199. Another signaling pathway, the signaling of the hepatocyte growth factor receptor (also known as MET), was enriched in both the luminal A and Her2 subnetworks. Interestingly, MET downregulation increases ERBB2 (also known as Her2) activation200, which is the defining characteristic of the Her2 subtype. Furthermore, low expression of MET is seen in both the luminal A and Her2 subtypes201. Finally, although the ErbB signaling pathway was not enriched in the Her2 subnetwork, ERBB2 was one of the most connected nodes. The basal subtype is compelling because it has the worst prognosis and is enriched with tumors lacking expression of hormone receptors or ERBB2, which results in the fewest treatment options202. The basal driver subnetwork was enriched for genes related to the DNA damage response and this subtype is known to have defects in DNA repair203. The phosphoproteomic data indicated high phosphorylation levels of several proteins in this pathway and phosphorylation of substrates of the known DNA repair kinases ATM and PRKDC was upregulated. This might indicate a compensatory mechanism of the kinases in this pathway to regulate DNA damage. The other pathway enriched in the basal subnetwork was the alpha 6 beta 4 signaling pathway. The alpha 6 beta 4 integrin is highly upregulated in basal breast cancer204, so mutations in this pathway might be driving integrin expression. Finally, an interesting area of future exploration for the basal subtype is the ESR2 pathway. ESR2 was uniquely present in the basal driver subnetwork and any of its protein-interaction partners were highly phosphorylated. ERa has known functions in breast cancer and it is rarely expressed in basal breast cancer, unlike in the other subtypes. Much less is known about the function of ERb in breast cancer, although reports suggest it represses EMT. Since some basal breast cancers have expression of ERb, treatment with an ERb agonist might have a positive effect. The main limitation of this study was validation and phosphoproteomic significance, likely due to sample size. Only 12 Her2 samples out of the total 77 had phosphoproteomic data, which likely limited the significance. Increasing the sample size, better phosphoproteomic resolution, and comparison to normal samples might produce a better signal-to-noise ratio. Furthermore, the heterogeneity within a subtype might contribute to the overall noise and reduce the signal that can be extracted. Exploring sample-specific driver subnetworks might ameliorate this problem. Finally, it is important to note that this analysis was performed comparing one subtype relative to others. While a particular pathway might be downregulated in one subtype compared to the others, it may still be overactive compared to normal and could then still be a target for therapy.

47 CHAPTER 5

PHOSPHOPROTEOMIC DATA REVEAL A DUAL ROLE FOR RB1 IN COLON CANCER

Introduction Colorectal cancer is one of the leading causes of cancer and cancer-related deaths in the United States205. It consists primarily of adenocarcinomas arising from colon and rectal epithelial cells. Usual treatments include surgery for early stage cancer. For more advanced tumors, chemotherapy with or without targeted therapy is often used in combination with surgery. Chemotherapy consists of some combination of leucovorin, 5- fluorouracil, oxaliplatin, irinotecan, and capecitabine, which all disrupt proper cell division. Additionally, targeted anti-VEGF therapies, such as bevacizumab or aflibercept, or anti- EGFR therapies, such as cetuximab or panitumumab, are used206. While the survival rate for colon cancer is high for early stages (> 90% five-year survival), patients with later stages still have poor survival and few targeted therapy options205. Global genomic studies on colorectal cancer tried to find new targets by providing frequently mutated genes, identification of fusion genes, and alterations in transcription factor activity207. However, these targets have yet to become clinically relevant, partly because the effect of many mutations on protein functions are not yet known. Proteins are the primary targets in cancer and therefore studying global changes at the protein level should help identify new therapeutic targets. Previously, CPTAC generated proteomic and phosphoproteomic datasets for three cancer types. They then extended the colorectal cancer study by generating data for a new cohort of colon cancer patients with the addition of phosphoproteomic data and matched normal samples for both the proteomic and phosphoproteomic datasets. This allows for removal of baseline protein expression and shows alterations specific to tumor samples. This new phosphoproteomic data allows colon cancer kinase signaling to be probed. As in most cancers, kinase signaling in colon cancer is dysregulated. Ras and BRAF are both kinases that are highly mutated in colon cancer208. Together with upregulation of EGFR, these mutations result in activation of the MAPK pathway209. There are also mutations in the PI3K pathway and SRC is known to be overexpressed210,211. Furthermore, we used the phosphoproteomic data to explore the effect of phosphorylation on substrates and clarify a contradiction in the colon cancer literature. The tumor suppressor gene retinoblastoma 1 (RB1) produces a protein that is a master cell cycle regulator. In normal cells, RB1 is monophosphorylated and bound to the transcription factor E2F212. Upon phosphorylation by members of the CDK family of kinases, RB1 releases E2F to drive transcription of cell-cycle related genes. In many cancers, RB1 is mutated or deleted, supporting its role in cancer proliferation213–215. However, RB1 is rarely mutated in colon cancer and it has been reported to be amplified in about 30% of colon cancer tumors with minimal deletion in other samples216,217. Additionally, RB1 mRNA expression and protein was increased in more than 80% of operable colorectal cancer cases218. Meanwhile in normal tissue, RB1

48 expression is confined to a small set of epithelial cells in the transition zone219. Furthermore, using western blot, several groups also showed that RB1 is phosphorylated in tumor tissue compared to normal220,221. We hypothesized that RB1 might be inactivated by phosphorylation rather than mutation or deletion in colorectal cancer. We used phosphoproteomic data to explore this idea.

Methods Data Data were obtained from the CPTAC consortium. The data consisted of 197 individual samples, which included 96 matched tumor-normal colon adenocarcinoma pairs, with copy number, TMT global proteomic, and TMT phosphoproteomic data. Six hundred eighty-eight human kinases and 255 phosphatases were defined as in Chapter 2.

Processing Phosphoproteomic Data Phosphopeptide identification was performed using MS-GF+222,223 to match against the RefSeq database (version April 2017). Site localization was performed using the Ascore algorithm224. Identified peptides were mapped to UniProt sequences (version July 2017) and named according to the canonical UniProt sequence. If the peptide matched multiple canonical UniProt sequences, the best ID was chosen based on presence of the protein in the proteomic data. If no canonical IDs had proteomic data, or if more than one protein was present in the quantified proteomics data, an ID was chosen at random. For peptides not matching a canonical protein sequence, a matching protein isoform ID was chosen. Peptides were filtered to those with an Ascore of ≥ 19 in at least one scan and a Q value < 0.01. Peptide abundances were log2 transformed and zero- centered for each gene. Data were re-centered across all samples to achieve a common median of 0. Phosphorylation site levels were determined by the median level for all peptides matching that site. Quantified sites and proteins were defined as those containing non-missing values in at least 50% of the matched samples. Data quality was assessed using principal component analysis (PCA) with the prcomp function in R.

Tumor vs Normal Analysis Log fold change was calculated as the log2 peptide ratios for normal samples subtracted from the log2 peptide ratios for tumor samples. Differential expression was performed using the paired Wilcoxon rank sum test. P values were adjusted using the Benjamini-Hochberg correction and an FDR < 0.05 was deemed significant. Phosphorylation site tumor markers were defined as sites upregulated more than 2x fold with an FDR < 0.01.

Term and Pathway Enrichment Overrepresentation analysis to describe proteins with upregulated and downregulated sites was performed using WebGestaltR with default parameters (minimum overlap 10, maximum 500). The background was the full list of proteins with phosphorylation data passing the same filters. The database for enrichment was Gene

49 Ontology Biological Process (GO BP). Results were considered significant with FDR < 0.05.

Biomarker Comparison Proteins containing phosphorylation sites upregulated > 2x fold were compared with protein biomarkers (upregulated > 2x fold and FDR < 0.01 in tumor compared to normal) and genes identified as having mutations and activity relevant to cancer in the Cancer Gene Census (CGC)225. Proteome data were quantified using unshared peptides and had non-missing values in at least 50% of tumor-normal samples.

Kinase Activity Prediction Using GSEA Unique phosphorylation sites were identified as a thirteenmer sequence. Phosphorylation sites of kinases were determined by a union of kinase-substrate interactions in PhosphoSitePlus (version May 2018), HPRD (version 9.0), and Swiss-Prot (version May 2018). The median log2 fold change of sites with at least 50% non-missing values was used to rank the phosphorylation sites and was submitted to the pre-ranked GSEA function in WebGestaltR. A minimum set size of 3 substrates was required and 1000 permutations was used to determine significance. Kinases were considered over- active with NES > 0 and FDR < 0.05. Kinases were considered under-active with an NES < 0 and FDR < 0.05.

Kinase Activity Prediction Using Regulatory Sites Regulatory sites with known effect on kinase activity were downloaded from Signor. Differentially upregulated sites in tumor vs normal (FDR < 0.05) that were known to activate kinase activity were used to predict increased activity and differentially downregulated activating sites were used to predict decreased activity.

RB1 Characteristics RB1 CNA and protein data were downloaded from CPTAC. The lollipop plot diagram was created using the R package trackViewer226. ssGSEA Activity Scores Protein activity in individual samples was determined using ssGSEA in the GSVA package in R. Substrate sets for kinases from the GSEA method were used to determine enrichment of phosphorylation sites of kinases. Substrate sets for E2F were downloaded from ENCODE227 and Hallmark sets were downloaded from MSigDB228. For enrichment of E2F targets and the Hallmark pathways, log2 fold change of proteins was used as the ranking method. Ten substrates were required for each analysis.

Correlation of RB1 Features to Cell Line Drug Sensitivity mRNA expression and cell line drug response were downloaded from the Genomics of Drug Sensitivity in Cancer (GDSC) project229. Response to CDK inhibitors was compared between cell lines with the lowest mRNA expression of RB1 (expression < 3) and the highest mRNA expression (expression > 6) using the student’s t-test.

50 Reverse phase protein array (RPPA) phosphorylation data for a subset of these cell lines were downloaded from the MD Anderson Cell Lines Project (MCLP)230. RB1 phosphorylation at S807/S811 was correlated with CDK inhibitor response using Pearson correlation.

Results Phosphorylation Changes in Colon Cancer The colon cancer dataset consisted of 197 samples (96 tumor samples with matched-normal tissue) with phosphoproteomic data. There were 71,504 unique identified peptides in the mass spectrometry data (Table 10). After filtering for high quality peptides, 31,339 unique phosphorylation sites could be identified. Of these sites, 7,298 sites on over 2,500 proteins were present in at least 50% of the matched tumor-normal pairs. Most of these sites (98%) have been identified in other experiments and reported in PhosphoSitePlus.

Feature Number Identified peptides 71,504 Filtered peptides 42,790 Phosphorylation sites 31,339 50% paired non-missing 7,295 Phosphorylated serine 6,444 Phosphorylated threonine 786 Phosphorylated tyrosine 65 Table 10. Colon phosphoproteomic data by the numbers.

The phosphorylation patterns differed between tumor and normal samples. PCA analysis showed good discrimination between the two tissue types in the first component (Figure 19A). Additionally, median phosphorylation levels across all of the sites were lower in tumor samples compared to normal (Figure 19B).

51

A B ●

60 ● ● ● ● Tumor ● ● Normal ● ● ● 2 ● ● ● ● ● 40 ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● 0 ● ● ● 20 ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● 0 ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ●● ●● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● − 2 ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

− 20 ● ● ● ● ● ● ● Median Log2 TMT Ratios ● ● ●

● − 4 PC2 (13% Variance Explained) PC2 (13% Variance ● ● ● ● ● − 40 −40 −20 0 20 40 Normal Tumor

PC1 (26% Variance Explained)

Figure 19. Difference between phosphorylation in colon tumor and normal samples. A) PCA for tumor (red) and normal (black) samples. B) Median log2 phosphorylation site levels in normal and tumor samples.

To assess changes in phosphorylation levels in individual tumors compared to normal tissue, the log2 peptide ratio for normal samples was subtracted from the log2 peptide ratio for cancer samples. Out of the 7,298 quantified sites, 5,895 of these sites also had TMT mass spectrometry protein data. The tumor log fold change in protein abundance and the tumor log fold change in phosphorylation levels were highly correlated (Figure 20A, Pearson correlation = 0.81, p value < 2.2x10-16). Using the signed rank Wilcoxon test, differential abundance of phosphorylation sites was determined. Out of the 7,298 quantified sites, 2,363 sites were significantly upregulated in tumor compared to normal, while 3,338 were downregulated (Figure 20B). Proteins with phosphorylation site abundance higher in normal were enriched for stromal-related proteins. The top 10 enriched terms for downregulated sites in tumor were all related to cytoskeleton organization and cell locomotion (Appendix E, Table E1, FDR < 3x10-10 for all 10 terms). This indicated that normal samples likely contained a higher percentage of stromal and muscle cells, while tumor sample composition was primarily epithelial cells. Top enrichment terms for proteins with upregulated sites included RNA processing and splicing and organization (Appendix E, Table E2, FDR < 5x10-12). There was no enrichment for the 42 proteins with hypophosphorylated sites (indicated in blue in Figure 20A), but the 62 proteins with hyperphosphorylated sites (indicated in red in Figure 20A) were related to muscle structure and contraction and cytoskeleton.

52 ● ● ● ● ●●●●● ● ●● ●● ●●●● ●●●●● ● ● ●●● ●●●●● ●● ●● ●● ●●●●●● ●●● ● ● ●● ● ●●● ● ● ● ● ● ● ● ●●●● ● ●●● ● ●●● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ●●● ●● ●●● ● ● ●● ● ● ● ● ● ● ●● ●● ●● ● ● ● ● ● ● ●●● ●● ●●●●● ●● ●●● ● ● ● ● ● ● ● ● ●●● ● ●●● ●●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ●● ● ●● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ●●●● ● ● ●● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●● ● ● ● ● ● ● ● ● ● ● ●●● ● ●●● ●● ● ● ●● ● ● ● ● ● ● ● ● ●● ●● ●● ● ●● ●●● ●● ●● ● ● ● ●● ●● ● ●● ● ●● ●● ●●● ●●●● ●●●●● ●● ●●●●●●● ● ●● ●● ● ● ● ● ● ● ●● ●●● ●● ●●●● ●●●●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●● ●● ● ● ●● ● ● ● ● ●●● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ●● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ●● ● ● ● ● ●●●●● ● ● ● ●● ● ●● ● ● ●● ● ● ● ● ● ●●● ● ● ● ●●● ● ●●● ● ● ●●● ● ●● ● ● ● ●● ● ● ● ● ● ●● ● ● ●● ●● ● ● ● ● ● ● ● ● ●● ● ●● ● ●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ●●●●● ● ● ● ● ● ● ● ●● ●● ●● ● ● ● ● ●● ●● ● ● ●● ● ● ●● ●● ● ● ●●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ●● ● ●● ● ● ● ●● ● ●●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ●● ●● ● ● ● ●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ●● ● ● ● ●● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ●● ● ●● ● ● ● ● ● ● ● ● ● ●●● ● ●●● ● ● ●● ● ● ●●● ● ● ●●●● ● ● ● ● ●●● ●●● ● ● ● ● ● ●● ●● ● ● ● ● ● ● ●● ● ● ● ● ●● ●● ● ● ● ● ●● ●●● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ●● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ●●● ● ● ●● ● ●● ● ● ● ● ● ● ●● ●● ●● ● ● ● ● ● ● ● ● ●● ● ●● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ●● ●● ● ● ● ●● ● ● ●●● ● ●● ● ● ●● ●● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ●● ●● ●● ● ● ● ● ●● ● ● ● ● ●● ● ● ●●● ●●●● ● ● ● ● ● ● ● ● ● ● ●● ● ● ●●● ●● ●●● ● ● ● ● ● ●● ● ● ● ● ●● ● ●● ● ● ● ● ● ● ● ● ●● ●● ● ● ● ●● ● ●● ● ●● ● ● ● ● ● ● ● ● ● ● ●● ●●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ●● ● ● ●● ●● ●● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ●●●●●● ● ●●● ● ●●● ● ● ● ●● ● ● ●● ●●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●●● ● ● ● ●● ● ● ● ● ● ● ● ●● ● ● ● ● ●●●● ● ● ● ● ● ● ● ● ● ●● ●● ● ● ●●●● ●● ●● ● ● ● ● ● ● ● ●●●●● ● ● ●●●● ● ● ● ● ● ● ●● ● ● ●● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●● ●●● ●● ● ● ● ● ● ● ●● ●●● ● ● ● ●● ● ●● ● ●●● ● ● ● ●●● ● ● ●● ● ● ● ●● ● ● ●● ● ●● ●● ● ● ● ●● ● ●● ● ● ● ● ● ●● ●● ● ● ● ●●● ●● ● ● ● ● ● ●● ● ●●● ● ● ●● ● ● ●● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ●●● ● ● ● ● ● ● ● ●●●●● ● ● ● ● ● ● ● ● ● ● ● ●● ●● ●●● ● ● ● ● ● ●● ● ● ●● ● ●● ● ●●● ● ● ● ● ● ● ● ● ● ●● ● ●● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ●●●● ●● ● ● ● ● ●● ● ● ● ● ●●● ●● ●● ● ● ● ● ● ● ● ●●● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ●● ●●● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ●●● ● ● ● ●●●● ● ●● ● ●● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ●● ●● ● ● ● ● ● ● ●● ● ●● ●●● ● ● ● ● ● ●● ● ● ● ● ●● ●● ●●●● ● ● ● ● ● ●●● ● ●● ● ● ● ● ● ● ● ●● ●● ●● ● ●●● ●● ● ● ●●● ●● ● ● ● ● ●● ● ●●●● ● ●● ● ● ● ●● ● ● ● ● ● ● ● ● ● ●● ●● ● ●●● ●● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ●●● ● ● ●●● ●● ● ● ● ● ●●● ● ●● ● ●●●● ● ● ● ●●● ●● ● ● ● ● ● ●● ● ●● ● ●● ●● ●● ●● ●● ●● ● ● ● ●● ● ● ● ●● ●●● ●● ● ● ● ● ● ● ●● ● ● ● ● ●● ●●● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ●●● ● ● ● ● ● ● ● ● ●● ● ●● ●●●● ● ● ● ●● ● ●● ● ● ● ●● ●●● ● ● ●● ● ● ●● ● ●● ●● ●● ●●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ●● ●●● ● ● ●●● ● ● ● ● ● ● ● ● ● ●● ● ●●●● ● ●●● ● ●● ● ● ● ● ● ● ● ● ● ● ● ●● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ●● ●● ● ● ● ●● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ●● ●●● ● ● ● ●● ● ●● ● ● ● ● ● ● ● ● ●●● ●●●●●● ● ● ● ●● ● ● ● ● ● ●●● ● ●●●●● ●● ● ● ● ● ● ● ●● ●● ● ●● ● ● ● ● ● ●●●● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ●●● ●● ●● ● ●●●● ● ● ● ●●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●● ● ● ●●●●●● ● ● ● ● ● ●●●● ●● ● ● ● ●● ● ●●●●● ● ●●●●● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ●● ● ● ● ● ●● ● ● ● ● ● ● ● ●● ●● ● ● ● ● ● ●●● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ●● ● ● ● ● ●● ● ●● ●●● ● ● ●●● ●● ●●● ● ● ● ● ● ●●● ●● ● ● ● ●● ● ● ●● ● ●● ● ● ● ● ● ● ● ● ● ● ●● ●● ● ● ●●● ●●●●● ● ● ● ● ● ● ● ● ● ● ●●● ● ● ● ●●● ● ● ●● ●● ● ● ●●●● ● ● ● ● ● ●● ● ● ● ●● ●●● ● ● ●● ●● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ●● ● ●● ● ● ●●● ● ● ● ● ● ● ● ● ● ● ● ●● ● ●●● ● ● ●●●● ●●● ● ● ● ● ● ●● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ●●● ● ●● ● ● ●●●●●●● ● ● ● ● ●● ●● ● ● ● ● ● ● ●● ● ● ●●● ●●● ● ● ●● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ●●● ● ● ● ●● ● ● ● ● ●● ● ● ●● ● ● ●● ● ● ●● ● ● ● ● ● ● ● ●● ●● ● ● ● ● ● ● ●● ●● ● ● ● ● ●● ● ● ● ●●●● ● ● ● ● ● ● ● ●● ● ● ● ● ●● ● ● ● ●● ● ● ● ● ● ● ● ●●● ● ● ● ● ● ●●●●● ● ● ●●● ● ● ● ● ● ● ● ● ●● ● ●●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●● ● ● ● ● ● ● ● ● ● ● ●● ● ● ●● ● ●●●● ●● ● ●● ● ● ●● ●●● ● ● ● ●● ●● ●● ● ●●● ● ●●● ● ● ●●● ● ● ● ● ●● ●●● ● ●●● ● ● ● ● ● ● ●● ●● ● ● ● ● ● ● ●●● ● ● ● ●● ● ● ●● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ●● ● ●● ● ● ● ● ● ●● ● ●●●●● ● ● ●● ● ● ● ● ● ●● ● ● ●● ●● ●●● ● ●●● ● ●● ● ● ●● ● ● ● ●● ●● ● ● ● ● ● ● ●● ● ●●●●● ● ●● ●● ● ●● ●● ●●● ● ● ● ● ● ● ● ● ● ●●● ●●● ● ● ● ●● ●● ● ● ● ● ● ●● ●● ●● ●● ● ● ●● ● ● ● ● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●●● ● ● ● ● ● ● ● ●●● ●●●●●● ● ● ● ● ● ●● ●● ● ● ● ●● ● ● ● ● ● ● ● ●● ● -log10(FDR)) ● ● ● ● ● ● ● ● ●●●●● ● ● ● ● ● ● ● ● ●● ● ● ● ●● ● ●●● ● ● ● ●●● ●●●● ● ●● ● ● ● ● ●● ● ●●● ● ● ● ● ● ● ●● ●●● ● ●●●●● ● ● ● ●● ●● ● ● ● ●●● ● ● ● ● ●●● ● ● ● ●●●● ● ● ● ● ● ●● ● ●● ● ● ●●● ●●●●●● ● ● ● ● ● ● ● ● ●● ●● ● ● ●● ● ●● ● ● ●● ● ● ●● ● ● ● ● ● ● ● ● ● ● ●●●● ●● ● ● ●● ● ●●●●● ● ●●● ●● ●● ●● ●●● ● ● ● ● ●●●●●●●●●●●●●●● ● ● ● ● ● ● ●●● ● ●●● ● ● ●●●● ●●● ● ●● ● ●●●● ● ● ● ● ● ● ● ● ● ● ● ● ●●●● ● ● ● ● ● ● ● ● ● ●●●● ●● ●● ●●●●●●●●● ●● ● ● ● ● ● ● ●● ●●●● ● ● ●●●●●●● ●● ● ● ● ● ●●●●● ● ● ● ● ●●●● ● ● ● ● ● ●●●●●●●●● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ●●●● ● ● ● ● ● ●● ● ● ● ● ●● ●● ● ● ●●●●● ● ● ● ● ● ● ● ●● ● ●●●●●●●●●● ● ● ● ●● ● ● ●●●●●●● ● ●● ● ● ● ● ● ● ● ● ● ●● ●●● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●●●●●●●● ●●●● ● ● ● ●●● ● ● ● ● ● ● ● ● ●● ●●●●●●● ● ● ● ●●● ● ● ●● ● ● ●●●●●●●●●●●●● ● ● ● ● ● ● ● ●●●● ●● ● ●● ●●●● ● ● ● ●● ● ● ● ●●●●●●●●●●● ● ● ● ● ● ● ●● ● ● ●● ● ●● ●● ●● ● ● ●●●●●●●●●● ● ● ● ● ● ●● ●● ● ● ● ●●● ● ●● ●●●● ● ●● ●● ● ● ● ● ● ● ● ● ●●●●● ● ● ●●● ● ● ● ●●● ●●●●●●●●●● ● ● ● ● ● ● ●● ● ● ●● ●● ● ●●●● ● ● ● ●●●●● ● ● ● ● ● ● ● ●● ● ● ● ● ●●●●●●●●●●●● ● ● ● ● ● ●●● ● ● ●● ●●●●●●● ●●●●● ● ● ● ● ● ● ● ●●●● ●●● ● ● ● ● ●●●●●●●●●●●●● ●● ● ● ●● ●●● ●●●● ● ●●●●● ● ● ● ● ●●●●●●● ● ● ● ●● ● ●● ●●●●●●●●●●●●● ● ●●● ● ● ● ● ●● ● ●●● ●●●●●●●●●● ● ● ● ● ● ●● ● ●● ●● ● ● ●●●●●●●●●●●●●● ●●● ● ● ● ●● ● ● ● ●● ●● ● ● ● ● ● ● ● ●● ● ● ● ●●●●●● ●● ● ● ● ●● ● ● ● ● ●●● ( ● ● ● ●●●●●●●●●●●●●●●● ● ● ● ● ● ● ● ● ●●● ● ●● ● ● ● ●●●●●●●●●● ●●●●● ● ● ●● ● ● ●● ● ● ● ● ●● ● ● ● ●●●●●●●●●●●●● ● ● ●● ● ● ● ● ●●●●●●●●●●●●●●●●●● ● ● ●● ●● ● ●●● ●● ● ●●●●●●●●●●●●●●● ● ● ● ●●● ●● ● ● ● ●●●● ● ●●● ●● ●●● ●●●●●●●● ● ●●● ● ●● ● ● ●● ● ● ●●●●●● ●●●●●●●● ●●● ● ● ● ● ●● ● ●●● ● ● ● ● ●●●●●●●● ●●● ● ●●● ●● ● ● ●● ● ● ● ●●●●●●●●●● ● ● ●● ● ●● ● ● ●●●●●●●●●●● ●● ● ● ● ● ● ● ● ● ●●●●●● ●●●●●● ● ● ● ● ●● ● ● ● ●●●● ●●●●●●● ● ● ●● ●●●●●●●●●●● ●● ● ●● ● ● ● ● ● ● ●●●●●●●● ● ● ● ● ● ● ● ● ●●●●●●●●●● ●●●●●●●●●●● ● ● ●● ● ● ●●● ● ●● ● ● ● ● ●●●●●●●●●●●●●●●●●●● ● ● ● ● ● ● ● ● ● ●● ● ●● ● ● ● ● ●● ●●●●●●●●●●●● ● ● ● ● ● ● ● ●● ● ● ● ●● ●● ●● ● ●● ●●●●●●●●● ●● ● ● ● ●● ●● ●● ●●●●●●●●●●●●●●●●●●● ●● ● ●● ● ●●● ● ● ● ●●●● ●● ● ● ●●●●●●●●●● ● ● ● ●●● ● ● ●● ●● ●●●●●●●●●●●●●●●● ●● ● ● ● ●● ● ● ● ●● ●● ●●● ● ●●● ● ● ● ● ●●●●●●●●●●●● ●● ● ● ● ● ● ●● ● ●● ● ● ● ● ● ● ● ● ● ●●●●●●●●● ● ● ● ● ●● ●● ● ● ● ● ●●●●●●●●●●●●●● ●● ● ● ● ● ● ● ● ● ● ● ●● ●●●●●●●●● ● ● ● ● ● ● ● ● ● ● ●●●●●●●●●●●●●●●●●●● ● ● ● ● ● ●● ● ●●●● ● ●● ● ● ●●●●●●● ●● ● ●● ● ● ● ● ●●●●● ● ● ● ●●●●●●●●●●●●●●●●●●●●●●● ●● ● ● ● ●● ●● ● ●●● ● ● ● ● ● ●●●●●●●●●●●●● ● ●● ● ●●●● ● ● ● ● ●●●●●●●●●●● ●●●●●●●●●● ● ●● ●● ●● ● ●● ● ● ● ● ●●●●●●●●●●●●● ● ● ● ● ● ●● ●●● ●●●●● ● ● ●● ● ●● ● ● ● ●●●●●●●●●●●●●●●●●●●●●●● ● ● ●● ●● ●● ● ●●●● ●●●●●●●●●●●●●● ● ● ● ● ●●● ●● ● ● ● ● ●●●●●● ● ● ●●●●●●●●●● ● ●● ● ● ● ●● ●● ● ● ● ●●● ●●●●●●●●●●●●● ● ●● ● ● ● ●●●●●●●●●●●●●●● ● ● ●● ●● ● ● ● ● ● ●●●●●●●●●●●●●●●●● ● ● ●● ● ● ● ● ● ● ● ● ● ● ●●●●●●●●●●●●●●●● ● ● ● ●● ● ● ● ● ● ● ● ● ● ●●●●●●●●●●●●●●●●●●●●●●●●● ● ● ●● ●● ● ● ● ●● ● ●● ●●●●●●●●●●●●● ● ● ● ● ● ● ● ● ●●●●●●●●●●●●●●●● ● ● ●● ● ● ● ● ● ●● ●●●●●●●●●●●●● ● ● ● ● ● ● ● ●●●●●●●●●●●●●●●● ● ● ● ● ● ● ● ●● ●● ● ●●●●●●●●●●●●● ● ●● ● ●● ● ● ● ●●●●●●●●●●●●●● ● ● ● ● ●● ●● ● ●● ●● ●● ● ●● ● ●● ●●●●●●●●●●●●● ●● ● ● ● ● ● ● ●● ● ●● ● ●●●● ●●● ● ● ● ● ●●●●●●●●●●●●●●●●●●●● ● ● ● ● ● ●● ● ● ●●● ● ● ● ●● ●●●●●●●●●●●●●●● ●●●● ● ● ●● ●● ● ●●●●● ● ● ●●●●●●●●●●●●●●●●●● ● ● ● ● ● ● ●● ●●● ● ● ●●●●●●● ● ● ● ●●● ● ●● ● ● ●●●●●●●●●●●●● ● ● ● ●● ● ● ●● ● ●●●●●●●●●●●●●●●●●●●●●●● ● ● ●●● ● ● ●● ● ● ●● ●●● ● ● ● ● ●●●● ●●●●●●●●●●●●● ● ● ● ●● ● ● ●●●●●●●●●●●●●●● ● ● ● ●●●● ● ●● ●● ● ● ●●●●●●●● ●●●●●●●●●● ● ●●●●●● ● ●●● ●● ● ● ●●●●●●●●●●●●●●●● ● ●● ●● ● ●●●●●●●●●●●● ● ● ● ● ● ●● ●● ●●●●●●●●●●● ● ● ●●● ● ● ● ● ● ● ● ● ●●●●●●●●● ●●●● ● ● ● ● ● ●● ● ● ● ● ● ● ●●●●●●●●●●●● ●● ● ● ● ●● ●●● ● ●● ● ● ●●●●●●●●●●●●●●●●●● ● ●●● ●●● ● ● ● ● ● ●●●●●●●●●● ● ● ● ● ● ●●● ● ●●● ●● ●● ●●●●●●●●●●●●●●●●●●●●●●●●● ● ● ● ● ●● ● ●● ● ● ●●● ● ● ●●●●●●●●●●●●●●●●●●●●●●●● ● ● ● ●●● ● ●● ● ●● ●● ● ● ● ●●●●●●●●●●●●●●●●●●●●●● ● ● ●●● ● ● ● ●●●●●●●●●●●●●●●●●●●● ● ● ● ● ●● ● ● ● ● ● ● ● ● ●● ●● ●●●●●●●●●●●●●●● ● ● ●● ● ●●● ● ● ●● ●● ● ● ● ●● ●●●●●●●●●●●●●●●● ●● ● ● ●●●● ● ●● ●● ●● ●●● ●●●●●●●●●●●●●●●●● ● ● ● ● ●● ●● ● ● ●● ● ● ●●●●●●●●●●●●●●● ●●●● ● ● ● ● ●●●● ● ●● ● ● ●● ●●●●●●●●●●● ●●● ●●●●●●●●●●● ●●●● ● ● ● ●●● ● ● ● ● ● ●● ● ● ●● ● ●●●●● ●●●● ●● ●●● ●● ● ● ● ● ●●●●●●●●●●●●●●●● ●●●●●●●● ● ● ● ● ● ● ● ● ● ●●●●●●●●●●●●●●●●● ● ● ● ● ●●● ● ● ● ●● ● ●●●●●●●●●● ● ● ● ● ● ●● ●● ●●●●● ●●●● ● ● ● ●● ●● ● ● ●● ●● ●●●●●● ●●●●●●●● ● ● ● ● ● ● ● ● ● ●●●●●●●●●●● ● ● ●●●● ●● ●● ● ●●●●●●●●●●●●●●●●●●●●●●● ● ● ●● ●● ● ● ● ● ● ● ●●●●●●●●●●●●●●● ●●● ● ● ●●●●● ● ●●● ● ● ●●●●●●●●●●●●●● ● ● ● ● ● ●● ● ●● ● ●●●●●●●●●●●●●●● ● ● ● ●●●● ● ● ● ● ● ● ● ●● ●● ●●●●●●●●●●●●●●●●●●●● ● ● ● ●●●● ● ● ● ● ● ●● ●● ● ●●●●●●●●●●●● ●●● ●● ● ● ●● ●● ● ● ●●●●●●●●●●●●●●●●● ● ● ● ● ● ●●● ● ● ● ● ●●●●●●●●●●●●● ●●● ●●● ● ● ●● ● ● ●●● ●● ● ● ● ● ● ● ● ●●●●●●●●●● ●●●●● ● ● ● ● ●●●● ● ●●● ● ● ● ● ● ●●●●●●●●●●●●●●●●●●● ● ● ● ●● ● ●●● ●●●● ●● ● ● ●●●●●●●●●●● ●● ● ●●● ● ●● ● ●●● ● ● ●● ● ● ● ● ● ●●●●●●●●●●●●●● ● ● ● ● ● ●● ●● ● ● ●●●●●●●●●●● ● ● ● ● ● ●● ● ● ●● ● ● ● ● ●●●●●●●●●●●● ● ●●● ● ● ● ●● ●●●●●●●●●●●●●●●● ● ● ● ●● ●●●● ●●● ●● ● ●● ● ●●●●●●●●●●●● ● ● ● ●●●●● ● ●● ● ● ● ●●●●●●●●●●●●●●●●● ● ●●● ● ● ● ● ●●●●●●●●●●● ● ●● ● ● ● ●●●●● ●●● ● ● ● ● ● ●●●●●●●●●●●●●● ●● ● ● ● ● ● ● ● ● ● ● ●●●●●●●●●●● ●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ●● ● ●●●●●●●● ● ● ● ● ● ●● ●● ● ● ●● ● ● ● ● ●●●●●●●●●●●●●● ● ● ● ● ●● ● ●● ● ● ● ● ●●●●●●●●●●●● ● ● ● ● ● ●●●● ● ● ●● ● ● ●● ●● ●●●●●●●● ●● ●● ● ● ● ●● ● ●●●● ● ● ● ● ●●●●● ● ● ● ● ● ● ● ● ●● ● ● ● ●●●●●●●● ●● ● ● ● ● ● ●● ●●●● ●●●●●●●● ● ● ● ● ●●●●●●●●● ●● ● ● ● ● ● ● ●● ● ● ●●●●●●●●●●●●●● ●●●●●●●●●●●●●●●●●●●●●● ● ● ● ● ● ● ●● ● ● ● ●● ● ● ●● ●● ● ● ●●● ●●●●●●●●●●●●●●●●●●●●●●●●●●●● ● ● ● ●●●● ● ● ●● ● ● ● ●●●●●●●● ●● ● ● ● ●● ●● ● ●●●●●●● ●●●●●● ●● ●● ● ● ●●● ●●● ● ● ● ●●●●●●●● ● ● ●●● ●● ● ● ●● ●●● ●●●●●●●● ● ● ● ● ● ● ●●● ● ● ●● ●● ● ● ●●●●●●●●●●●●●●●●● ● ●● ● ● ●●●●● ● ● ●●● ●● ● ● ● ● ● ● ● ●●●●●●●●●●●●●●●●●● ● ● ● ● ● ● ● ●●● ● ● ● ● ● ●● ●●●●●●●●●●●●●●● ● ● ● ● ● ● ● ●● ● ● ● ●●●●●●●●●●●●●●●●● ● ● ●● ●● ●● ● ● ● ●●●●●●●●●●● ●● ● ● ●●● ● ● ●●● ● ● ● ●●● ●●● ●● ● ● ● ● ● ● ● ● ● ● ● ●●●●●●●●● ● ● ● ●●● ● ● ● ● ● ● ● ●●●●●●●●●● ● ●● ● ● ● ● ● ●● ● ●● ● ●●●● ● ● ● ● ●●● ● ●●● ● ● ●●●●● ● ● ●●● ● ●● ● ● ● ● ● ● ●●●●●●●● ● ●●● ●● ●● ●● ● ● ● ● ● ●●●●●● ●●●● ● ● ● ●●● ● ● ● ● ● ● ●● ●● ● ● ● ●●●● ●●●●● ●● ● ●●●●●● ● ●●● ● ●● ●●●● ●● ● ● ● ● ● ●●●●●● ● ● ●● ●●●● ● ●● ● ● ● ● ●● ●● ●●●●●● ● ● ● ● ●● ● ●● ● ●● ●● ●●● ●●●●●●●● ● ● ●● ●●● ● ● ● ● ● ● ●●●●●●●●●●● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ●● ●● ● ● ● ●●●●●● ● ● ●● ● ●● ●● ● ●● ●●●● ● ● ● ● ●●● ●● ●●●●● ● ● ● ● ●● ● ● ● ● ●● ● ● ● ●●●● ● ● ● ● ●●●●●●●● ● ● ●● ● ● ● ● ● ● ● ● ● ●●●● ●●●●● ●●●●●● ● ● ● ● ●● ●●● ●● ● ● ●● ● ● ● ● ● ●● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ●● ●●●● ●●● ● ● ● ● ● ● ● ● ● ●● ●●● ●● ● ● ● ● ● ● ● ●● ● ● ●● ●● ● ● ● ● ● ●●●● ●● ● ● ● ● ● ● ● ●●●●●● ● ●●● ● ● ●●●●●● ●● ● ● ● ● ● ●● ●● ● ● ● ● ● ●● ● ● ●● ●● ●● ●●● ●●● ● ●●●● ● ●●● ●●● ●● ● ●● ● ● ● ● ●●● ● ● ● ● ●● ●● ● ● ●●● ● ● ● ●● ●●● ● ●● ● ●●● ●● ● ●● ● ● ● ● ● ● ●● ●● ● ● ● ● ● ●●●●●●●●●● ●● ●●● ●●●● ●●●●●● ●●●●●●●●● ● ● ● ●● ● ●●● ●● ●● ● ●● ● ● ●●● ●● ● ● ● ● ● ● ●●● ●● ● ● ● ● ● ●● ● ● ●● ● ● ● ● ●● ●●●● ● ●● ●●● ● ● ● ● ● ● ●● ●● ●●● ● ● ●● ● ● ●●● ● ● ● ● ● ● ●●● ●●● ●● ● ● ● ●●● ●● ●● ● ● ● ● ● ●● ● ● ●●●● ● ●●●●●●● ● ● ● ●● ●●● ● ●●● ● ● ● ● ● ● ●● ● ● ●● ●●●●● ● ● ●● ●● ● ●●●● ●●● ● ● ●●● ●● ● ●● ● ● ●● ● ● ● ●●● ● ● ●●● ●● ●●● ●●●●●●●● ●●● ● ●● ● ● ●●●●● ●●● ●● ●●● ● ● ● ● ●●●●●●●●●●● ● ● ● ● ● ●● ● ● ● ● ● ● ●●●●●● ● ●● ●●● ● ● ● ●● ●● ● ●● ● ●●● ●●●● ● ●● ● ● ●● ● ●● ● ● ●● ● ● ●●●● ● ● ● ●●●●●● ●● ● ● ● ● ●● ● ● ● ● ●● ● ●●●● ●●● ● ● ● ●● ● ●● ● ●● ● ● ● ●●●● ● ●● ●●● ● ● ●● ●●●● ● ●● ● ● ● ●●●●●● ●●●● ●●●● ●●●● ● ● ● ● ● ●● ● ● ● ●● ● ● ● ● ● ●●●● ● ●● ●● ●● ● ● ● ●●● ● ● ● ● ● ● ● ●● ●● ●●●● ● ● ●● ● ● ● ● ●●● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●● ●●● ● ●●●● ●● ●●●● ● ● ● ● ● ●● ● ● ● ●● ● ●●●●●● ● ● ●● ● ● ● ●● ●● ●● ● ●●● ●● ●● ●● ● ●●●● ● ● ● ●● ● ● ● ● ●● ● ● ● ●●●● ●●● ●● ● ●● ● ●●●●●● ● ●● ● ● ●●●●● ●● ● ● ● ● ●●●● ● ● ● ●● ● ●●● ●● ●● ●● ● ● ●● ● ● ● ●●● ● ● ●●●● ●● ● ● ● ●● ●●● ● ● ● ● ●● ●●● ● ● ● ● ● ●● ● ● ●● ● ● ● ● ● ● ●●● ● ●● ●● ● ●●●● ● ●●● ●●● ● ● ● ● ● ●●●● ● ●●● ● ● ●●● ● ● ●●● ●● ● ● ●●● ● ● ● ● ● ● ● ●●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●● ●●● ● ● ●●● ● ● ● ● ● ● ● ● ●●● ●● ● ●● ● ● ●● ●● ● ● ● ● ●● ●● ● ●● ●● ● ● ● ● ● ● ●●●●● ● ●●●● ● ● ● ● ● ●●● ● ● ● ● ● ● ● ● ● ● ● ● ●● ●● ●●●● ●●● ● ● ●● ● ● ● ●●● ●● ●● ● ● ● ● ● ● ●● ●● ●● ●●●● ●● ● ●●●● ● ● ● ● ●● ● ● ●●●●● ● ● ●● ● ● ● ● ● ● ●● ● ● ● ●●● ●●●● ●● ● ● ● ●●● ● ● ●● ● ● ●● ● ● ● ●●● ●●● ●●● ●● ●● ●● ● ●● ●● ●● ●● ● ●● ● ●●●● ●●● ●● ●●●●● ●● ● ●●●● ● ● ●● ●● ● ● ●●● ● ● ● ●● ●● ● ●● ● ●● ● ● ● ● ● ● ● ●● ●● ● ● ● ● ● ● ●● ● ● ● ● ●●● ●●● ● ●● ● ● ●● ● ● ●● ● ● ● ● ●● ● ● ● ● ●● ● ● ●●● ● ● ●●●●●● ●●●● ● ● ● ●●● ● ● ●●● ● ● ● ● ●● ● ● ● ● ●● ● ●●●●●●● ●●●●● ●●● ●● ●●●●●● ●● ●● ● ● ●● ● ● ● ●●●● ● ●● ●●●● ● ●● ● ●●● ● ●● ● ● ●● ● ●● ●●● ●● ● ●●● ● ● ● ●● ● ● ● ● ● ● ● ● ● ●●●●●● ●● ● ● ●●● ●●● ● ● ● ● ●● ● ● ● ● ●● ●● ● ●● ● ●●●● ● ● ●● ● ● ● ●● ● ●● ● ● ● ●●● ● ● ● ●● ●● ● ● ● ●●● ● ● ● ● ● ● ● ● ● ● ●● ● ● ●●● ●● ● ● ● ● ● ●● ● ● ● ●● ●●● ● ●●● ●● ● ● ● ●●● ● ● ●●● ●●● ●●●●●● ●●● ● ●● ● ● ●● ● ●●● ● ●●●● ● ● ●●● ● ●● ●● ● ● ● ●● ● ●●●●● ● ●● ● ●●● ● ●● ●●●● ● ● ● ● ●● ● ●● ● ● ●● ● ●● ●● ● ●●●● ●● ● ●● ● ●● ● ● ●● ● ● ●●● ● ● ● ●● ● ● ● ● ● ● ● ● ●● ●●●● ● ● ● ● ● ●● ●● ● ● ● ●● ● ● ● ●● ● ● ● ●●●● ●● ● ●● ● ● ● ● ● ●● ● ●● ●●●● ●●● ● ●● ● ● ● ● ● ● ● ●● ● ● ● ●●● ● ●● ● ● ● ●●● ● ● ● ●● ●● ● ●● ● ● ● ●●● ●● ●● ● ● ●● ● ● ● ●● ●● ● ●● ●● ●● ● ● ● ● ● ● ● ● ● ●● ●●● ● ● ● ●● ●● ● ●● ●● ● ● ● ● ● ●●● ● ● ● ● ●●●● ● ●●● ● ● ●● ● ●● ● ● ●●● ●● ●●●● ● ●●●● ● ● ● ● ● ● ●● ● ●● ●●●● ● ●●●● ● ● ● ● ●●● ●●●●●●● ●●●● ● ● ● ● ●● ● ● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ●● ● ● ●● ●● ●● ● ● ● ● ●● ● ● ●● ●●●● ● ●● ● ● ● ● ● ● ● ● ● ● ● ●●● ● ●●● ●●● ● ● ●● ● ● ●● ● ● ● ●● ●●● ● ● ● ●●●● ●●●● ●● ● ● ● ● ● ● ●●● ● ● ●● ●● ●●●● ●● ● ● ● ● ●●● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ●●●● ●●●● ●●●●● ●●●●●● ● ● ● ● ●● ● ● ●●● ● ● ●● ● ● ● ● ●● ●● ●● ●●●●●●● ● ● ●● ● ●●●● ● ● ● ●● ● ●●●●●●●● ●● ●●● ●●●● ● ● ● ● ● ● ● ●● ●●●● ● ● ● ●● ● ●● ● ● ●● ● ●● ● ● ● ● ● ● ● ●● ● ● ● ● ●● ● ●● ●● ●● ● ●● ● ● ● ● ● ● ● ● ● ● ● ●●●●● ● ● ● ● ●● ● ● ●●● ● ● ●●●●● ● ●●●● ●● ●● ● ● ● ●●● ● ●●●● ● ●● ●● ● ●● ●●● ●● ●●● ●●● ● ●● ● ●● ● ● ● ●● ●●●● ● ● ● ● ● ●● ● ● ● ● ● ● ●●●● ●● ●●● ● ●●●● ● ●● ●● ● ● ●●● ●● ●● ●● ● ● ● ● ● ● ●● ● ●● ● ● ● ● ●● ●●● ● ●●● ● ● ● ● ● ● ●●● ● ● ●●●●● ●●● ●● ● ●●●● ● ●●● ● ● ● ●●●●● ● ● ●● ●●●●●●● ● ●● ● ● ● ● ● ● ●●●● ●●●●●● ● ● ●●●●●●● ●● ●● ● ●● ● ● ● ● ● ●●●●●● ●● ●●●●● ● ● ● ● ● ● ● ● ●● ●● ●● ● ●●●● ● ● ● ● ● ● ● ●●●● ● ●● ●●● ●● ● ● ● ●● ● ●●●●● ● ●● ●●●●● ● ● ● ● ● ● ●●●●●●●● ● ●● ●●●● ● ● ●● ● ●●●●●●●●●●● ●●● ●●●● ● ● ● ● ● ● ● ● ●●●● ●●●●● ● ● ● ●● ●● ●● ● ●● ● ●●●●● ● ● ● ● ●● ●● ● ●●●●● ●● ● ● ● ● ● ● ●●● ● ●● ● ● ● ● ● ● ● ● ●● ●● ● ●● ● ●● ● ●● ●●●●●● ● ●● ● ● ● ●●●● ●● ● ● ●●●●●●● ●● ● ● ● ●●● ● ● ● ● ● ●●●●● ● ●●● ● ● ● ●● ● ● ●●● ● ●● ●● ●● ● ● ● ● ● ● ● ●● ●● ● ● ●● ● ● ● ● ● ● ● ● ● ●●● ● ●●● ●●●●● ● ● ● ● ●●●● ●● ●●● ●●● ●● ● ●● ●● ●●● ●● ●● ● ● ● ● ● ●●●● ● ●●● ● ●● ●●● ●● ●● ● ●● ● ● ● ● ● ●●● ●●● ●● ●●●●●●●●● ● ● ● ●●●●●●●●●●● ● ● ●●●● ● ● ● ● ● ●● ● ● ●●●●●●●● ●●●● ● ● ● ● ● ● ● ●● ● ●●●●● ●●●●● ● ● ● ● ● ● ● ●● ●●●●● ● ●●● ● ● ● ● ● ● ●●● ● ● ●●● ●● ● ● ● ● ● ● ●● ● ● ●● ● ● ● ● ●●●●●● ● ●●● ●● ● ● ● ● ● ● ●● ●●●● ●●●●●● ●● ● ●● ● ●●●●●●●● ● ●● ● ● ● ● ● ● ● ●●●●●●●●●●● ●●● ●●●● ● ● ● ●● ● ●●● ●● ●●● ●● ● ● ● ● ●●●●●●●● ●●● ● ●● ● ● ● ● ●● ● ● ●●● ● ● ●●● ● ● ●●● ●● ●●●●●●●●●●● ● ●● ● ● ● ● ●●● ●●●●●●●●●●● ● ● ●●●●●●●●● ●● ● ●●●●●●●● ● ● ●●● ● ● ● ●●●●● ●●●● ● ●●● ● ●● ● ● ● ● ● ●●●●● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ●●●●●● ●●● ●● ● ● ● ● ●● ●●●●●●● ● ● ● ● ● ● ●● ● ● ● ●● ●● ●● ●●● ● ●● ● ● ● ● ●●●● ● ●●●●● ● ● ● ● ● ●●● ● ●●● ●●● ●●● ● ● ● ● ●●● ●●●●● ● ● ● ●● ●● ●●● ●● ● ●●●●● ● ● ● ●●●●●●●●●●● ● ●●●●●●● ●● ● ● ● ●●●●● ● ●●●●● ●●● ●● ● ● Phosphoproteome ● ● ● ●● ● ● ●●●● ● ●● ● ● ● ●● ●●●●●●●● ● ●●●● ●● ●●● ● ●● ● ● ●●●●●●● ●●●● ●●●●●● ● ● ● ●●● ●● ● ●●●●●●● ● ● ● ●●● ● ● ●●● ●● ● ● ● ● ●●●●● ● ● ●●●●●●● ●●● ● ● ● ● ●● ● ● ● ●●●●●● ●●● ● ● ● ● ● ● ●● ●● ● ●●● ● ● ● ● ● ●● ●● ●●●● ● ●●●●●●●●● ● ● ● ● ● ●●● ● ● ● ● ● ● ● ●● ● ●●● ●●●● ●●● ●● ● ● ● ● ● ● ●●● ●●● ● ●●●●●●● ● ● ● ●●● ● ● ●● ● ● ● ● ● ● ● ● ●●●●●●●● ●● ● ●●● ● ● ●● ● ●●●●● ●●●●● ●●●● ●●●●● ● ● ●● ●● ●●●●●●● ●●●●●●●● ●●●● ● ● ●●●●●●●●● ● ●● ●●●● ● ● ● ● ● ●●●●● ●●●● ●●● ● ●● ● ● ● ● ●●● ● ●● ●● ●● ●●● ● ● ● ●●●●● ●● ●●●● ●●● ●●● ● ● ● ● ● ● ●● ●●●●●● ● ● ●●●●●●● ●●●●● ● ● ● ● ●● ●●●●●●●●● ●●●● ● ● ●● ●●●●●●●● ● ● ● ● ● ● ● ● ● ●● ●● ● ●● ●●●● ●● ● ● ●●●●●●●● ● ● ● ●●●● ●● ●● ● ● ●●●●●●●● ●● ●●● ●●●●●● ●● ● ● ●● ● ●● ●●●●● ● ● ●●● ●●● ●● ● ● ● ● ● ● ●●● ●● ● ●●●●● ● ● ● ● ●● ●●●●●●●●●● ● ●●●● ●●● ● ● ● ●● ● ●●●●●●●●●● ●● ●●● ●●● ●● ● ● ●● ● ●●●●● ●●●●●●● ● ●●●●●●●● ●●● ●● ● ● ● ●●● ●●●●●●●● ●●●●●●●●●●● ●● ● ●● ● ● ●● ● ●●●●●●●● ● ● ● ●●●●●●●● ●●●●●● ● ●● ● ● ●● ●●●●●●●●●● ●●●●●●● ●● ●● ● ● ● ●●● ●●●●●●●● ●●●●●●●●●● ●● ● ● ●●●●●●●●●●●●● ● ●●●●●●●●●●●● ● ● ●●●●●●●●●●● ●●●●●●●● ● ● ●● ●● ●●●●●●●●●●● ●● ●●●●● ●●● ● ● ● ● ● ●● ●●●●●●●●●●●●● ●●●●●●●●●●● ●●● ● ● ● ● ●●●● ● ●●●● ●● ●●●●●● ●● ● ● ● ● ● ● ● ●● ●● ● ●●● ●●● ● ●● ● ● ● ●●●●●●●●● ●●●●●●●●●●●● ● ● ● ●● ●● ●●●●●●●●●●●● ●●●●●● ●●● ● ● ●● ●●●● ●●●●●●●●● ●● ●● ●●●● ●●● ● ● ●● ●●●●●●●●●●●●● ●●● ●●●●●● ● ● ● ● ● ●●●●●●●● ● ●●●●●●● ●●●●●●● ● ● ●● ●● ● ●● ● ● ● ● ●● ●●●●●●● ● ●●●●●●● ● ● ● ● ●●●●●●●●● ●●●●●● ●●●●●● ● ● ● ● ●●●●●●●●●● ● ●●●● ●●●● ● ● ● ●● ●●●●●●●●● ●●●●●●●●● ●●●● ● ● ● ● ●●●● ● ● ● ● ● ● ● ●●●●●●●●● ● ●●●●● ●● ● ● ●●●●●●● ●●●●●● ●●●● ● ● ●● ●●●●●●●●●●● ●●●●●●● ●●●●●●●●●●●●●● ● ● ●●● ●● ● ● ● ● ●●●● ●●●●●●●●●●●●● ●●●●●●●●●● ●● ●● ●●●●●●●●●●●●● ● ● ●●●●●●●● ● ● ● ● ● ●●●●● ●●●● ●●● ● ● ● ● ●● ●●●●●● ●●● ●●●●●●●●●● ●●● ● ● ●●●●●●●●●●●●●● ●●●●●●●●●●●●● ● ● ● ●● ●●●●●● ●●●● ● ● ● ● ●● ●●●●●●●●●● ● ●●●●●●●●● ●● ● ● ●●●●●●● ●●●●●●●●●●●●● ● Correlation = 0.81 ●● ●●●●●●●●●●●●●● ●●●●●●●● ●● ● ● ●●●●● ● ●● ● ● ●●●●●●● ●● ●●● ●●●●●●●● ● ●●●●●●●●● ● ●●● ● ●●●●●●●●●●● ●●●●●●●●●●● ● ● ● ●● ●●●● ●●●● ●●●●●●●●● ● ● ● ●●●●●●●●●●●● ●●●●●●●●●●●●●●● ● ●●●●●●●●●● ● ●●●●●●●● ● ● ● ●●●●●●●●●●●●●●●●●●●●●●●●● ●●● ● ● ●●●●●●●●●●● ●●●●●●●●● ● ● ●●●●●●●●●●●●●●● ●●●●●●●●●●● ● ●●●●●●●●●●●●● ●●●●●●●●●●● ● ●● ●●●● ●●●●●●●●●●●●●●●● ●●●● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●● ●●●●●●●●●●●●●●● ●●●●●●●●●● ●●●●●●●●●●●●●●●●●●●●●●● ● − 3 − 2− 1 0 1 2 3 ●●●●● ● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●● ●●●●●●●●●●●●●●●●●●●●●● ● ● ●●●●●●●●●●●●●●●●●●●●●●●●●● ● ● ●●●●●●●●●●●●●●●●●●●●●●●●●●● ● ● ●●●●●●●●●●●●●●●● ●●●●●●●●●●● ● ●●●●● ●●●●●●●●●●●●●●●●●●●●● ●●● ● ●●●●●●●●●●●●●●●●●●●●●●●●●● ●●●●●●●●●●●●●●●●●●●●●●●●● ●●●●● ●●●●●●● ● P−value < 2.2e−16 ●●●●●●●●●●●●●●●●●●●●●●●●●●● ●●●●●●●●●●●●●●● ● ●●●●●●●●●●●●●●●●●●●●●●● ● ●●●●●●●●●●●●●●●●●●●●● Signed Rank Test 0 2 4 6 8 10 12 14

−3−2−1 0 1 2 3 −3−2−1 0 1 2 3 Median Difference (log2FC, Tumor vs Normal) Median Difference (log2FC, Tumor Median Difference (log2FC, Tumor vs Normal) Median Difference (log2FC, Tumor vs Normal) Proteome Phosphoproteome

Figure 20. Characterization of phosphorylation sites in colon cancer. A) Median log2 tumor-normal fold change of proteins and sites on those proteins. Phosphorylation levels are highly correlated with protein abundance (Pearson correlation = 0.81, p < 2.2x10-16). The dashed red line shows the linear regression between abundances. Red colored points are sites 2x fold above the 45-degree line, while blue colored points are sites 2x fold below the 45-degree line. B) Differential expression of phosphorylation sites in tumor vs normal. Sites are colored based on significance (FDR < 0.01, blue = higher in normal, pink = higher in tumor. Darker color indicates absolute fold change ³ 2).

Comparison with Proteomic and Genomic Data There were 63 sites on 50 proteins with > 2x fold change and FDR < 0.01 in tumor samples compared to normal (Appendix F). Six genes identified in the CGC as important in cancer based on mutations had highly upregulated phosphorylation sites (DEK, NPM1, PML, RB1, TFRC, CDK12). Four proteins (DDX21, RSL1D1, S100A11, TOP2A) were highly abundant in tumor compared to normal and also had high phosphorylation site levels. No protein existed in all three, suggesting that each type of data added new information about proteins involved in colon cancer (Figure 21). Using GO enrichment analysis, the proteins with highly upregulated sites were primarily related to cell cycle and general DNA/RNA processes (Appendix E, Table E3). While the functions of most sites of these sites are unknown, a few highly upregulated sites have been studied in relation to cancer. Phosphorylation of the protein product of gene NOLC1 at site S563 inhibits the kinase CK2, which might prevent its pro- apoptotic function231. Topoisomerase II alpha (TOP2A) is upregulated in proliferating cells and controls the structure of DNA during the cell cycle232. Phosphorylation of TOP2A at S1106 enhances its activity233. Furthermore, phosphorylation of microtubule-associated protein 4 (MAP4) at S787 promotes its dissociation from tubulin, which results in depolymerization of microtubules234. Phosphorylation of this site further associates with resistance to the chemotherapy taxol235.

53

CGC

Proteome

1 26 692 0 4 6 40

Phosphoproteome

Figure 21. Overlap of cancer-associated genes and colon cancer-associated proteins and phosphoproteins. Number of proteins in the CGC, increased 2x fold in the protein data, or with sites increased 2x fold in the phosphoproteomic data.

Kinase Activity in Colon Cancer Kinase activity was inferred using enrichment of phosphorylation sites of specific kinases. Out of 78 kinase sets with at least 3 substrates in the colon data, 7 had increased activity and 7 had relatively decreased activity in tumor compared to normal (Figure 22A). The kinases with increased activity were primarily cell cycle related proteins (CDK1, CDK2, CDK3, CDK4, CDK6, CDK7, and CDC7). The kinases with relatively decreased activity in tumor were GSK3A and GSK3B, CDK5, MAPK12, DYRK1A, CK1, and PDPK1. Another way to infer kinase activity is to examine relative phosphorylation of regulatory sites on kinases and phosphatases. In the data, there were quantified sites on 185 kinases (160 of which were protein kinases) and 35 phosphatases. Sixty-one of the sites on kinases and 8 sites on phosphatases had a known effect on the enzyme. Of these sites, activating sites on 5 kinases were upregulated and activating sites on 31 kinases were downregulated in tumor (Figure 22B). Only 4 kinases (CDK7, GSK3A, GSK3B, and DYRK1A) were reported to be significantly over or under-active in colon cancer by both methods, while many of the remaining kinases were missing either phosphorylation or substrate information. Inhibitors for most of these kinases have undergone at least a phase I clinical trial, but only a few kinases are targeted by FDA- approved inhibitors (Figure 22B).

54 A CDK2 ●

3

PDPK1 GSK3A MAPK12 ● 2 ●● ●● CK1 CDK7 CDK4 CDK5 ● ● CDK3 ● CDC7 ● DYRK1A ●●● CDK1 CDK6 ● GSK3B ● ●● 1 ● ●● ● ● ● ●●● ● ●● ●●● ●● ●●● ● PreRanked GSEA ( − log10(FDR)) PreRanked ●●●● ●●●● ●● 0 ● ●●●●●●● −1 0 1 2 B Normalized Enrichment Score

Inferred Activity Phosphorylation Drug Target CDK5 CK1 MAPK12 PDPK1 PKN1 PKN2 RPS6KA4 PRKD1 PFKFB2 MTOR STK24 MAPK3 MAPK1 MAPKAPK2 MAP2K3 OXSR1 STK39 PRKCA PRKCZ ICK RPS6KA5 MAP2K1 MAP2K2 AKT1 DYRK1B DYRK1A GSK3B GSK3A MARK2 PAK4 RPS6KA1 RPS6KA3 PRKCB MAPK14 SRC CDK6 CDK1 CDK3 CDC7 CDK4 CDK2 CDK7 SRPK1 PFKFB3 PI4KB MELK

Figure 22. Kinase activity in colon tumor compared to normal. A) Kinase activity inferred using the pre- ranked GSEA method. The normalized substrate enrichment score is used as an activity score. Kinases with significantly different activity in tumor compared to normal are colored and named. Blue indicates relatively decreased activity, while red indicates increased activity. B) Summary of kinase activity using two different methods. Inferred activity are the significant kinases in (A). Phosphorylation indicates kinases with significantly increased (red) or decreased (blue) phosphorylation of their activating sites. White boxes indicate no significant difference. Gray boxes indicate kinase activity cannot be inferred for that kinase using that method. Kinases targeted by an FDA-approved inhibitor are indicated with black boxes.

RB1 Characteristics in Colon Cancer Because cell cycle kinases were highly active in the colon cancer samples and one of the proteins with multiple highly upregulated phosphorylation sites was the master cell cycle regulator RB1 (Appendix F), we hypothesized RB1 was being inactivated by phosphorylation instead of mutation or deletion. The RB1 gene was amplified in a majority of the 96 colon tumor samples (Figure 23A). In comparison to normal, RB1 protein was also increased in almost all samples (Figure 23B). Furthermore, eleven sites on RB1 were identified in the phosphoproteomic data and almost all were increased in tumor compared to normal (Figure 23D). Only six of these sites (S37, S249, T373, S807, S811, T826) were present in > 50% of the paired tumor-normal samples and five of these were

55 significantly increased in tumor compared to normal (Figure 23C). Unlike the other 5 sites, S37 is not known to affect RB1 activity so it was excluded from further analysis.

A RB1 CNA 15919 2 11CO030 05CO003 01CO005 11CO019 05CO026 09CO005 11CO008 01CO006 11CO005 01CO013 11CO033 05CO044 16CO002 11CO036 22CO006 01CO008 05CO039 11CO047 11CO007 16CO003 27CO004 21CO006 21CO007 11CO031 01CO014 05CO037 05CO050 09CO013 11CO021 11CO054 11CO061 05CO049 05CO033 05CO015 05CO048 11CO052 11CO018 22CO004 11CO057 11CO044 09CO006 11CO051 11CO042 01CO019 09CO022 11CO070 09CO019 05CO011 16CO006 05CO045 05CO034 05CO053 06CO002 11CO022 11CO072 20CO007 11CO058 11CO020 11CO045 20CO006 11CO027 11CO062 11CO037 20CO001 15CO001 06CO001 11CO048 20CO004 20CO003 11CO032 11CO079 11CO043 11CO010 09CO011 14CO005 05CO054 05CO032 11CO039 09CO018 05CO006 15CO002 05CO029 01CO022 05CO041 11CO053 05CO020 11CO077 05CO035 09CO008 09CO014 16CO011 05CO047 09CO015 05CO002 05CO028 01CO015 1 0 15919 2 −1 11CO030 05CO003 01CO005 11CO019 05CO026 09CO005 11CO008 01CO006 11CO005 01CO013 11CO033 05CO044 16CO002 11CO036 22CO006 01CO008 05CO039 11CO047 11CO007 16CO003 27CO004 21CO006 21CO007 11CO031 01CO014 05CO037 05CO050 09CO013 11CO021 11CO054 11CO061 05CO049 05CO033 05CO015 05CO048 11CO052 11CO018 22CO004 11CO057 11CO044 09CO006 11CO051 11CO042 01CO019 09CO022 11CO070 09CO019 05CO011 16CO006 05CO045 05CO034 05CO053 06CO002 11CO022 11CO072 20CO007 11CO058 11CO020 11CO045 20CO006 11CO027 11CO062 11CO037 20CO001 15CO001 06CO001 11CO048 20CO004 20CO003 11CO032 11CO079 11CO043 11CO010 09CO011 14CO005 05CO054 05CO032 11CO039 09CO018 05CO006 15CO002 05CO029 01CO022 05CO041 11CO053 05CO020 11CO077 05CO035 09CO008 09CO014 16CO011 05CO047 09CO015 05CO002 05CO028 01CO015 −2 1 B RB1 Protein RB1 Protein 0 −1 2.0 Normal

− −2 1.0 0.5

− RB1 Phosphorylation log2FC Tumor C RB1 Phosphorylation 3

●

2 ● Normal

− ● ● ●

1 ● ● ● ● ● ● ● 0 ● ● ● 1 − log2FC Tumor

D Normal - Median log2FC log2FC Median Tumor

Figure 23. RB1 characteristics in colon cancer. RB1 A) CNA, B) protein fold change, and C) phosphorylation fold change for five sites (S249, T373, S807, S811, T826) across all 96 samples. Samples are ordered by increasing average phosphorylation abundance. D) All identified phosphorylation sites on RB1. Color and height indicate median log2 fold change in tumor compared to normal.

56 RB1 Phosphorylation Correlates with Increased Proliferation and Decreased Apoptosis To determine the effect of RB1 phosphorylation, I correlated average RB1 phosphorylation abundance change with other data features. Average RB1 phosphorylation abundance change in tumor compared to normal was highly correlated with CDK2 activity at an individual sample level (Figure 24A, Pearson correlation = 0.54, p value = 1.5x10-8). It was also correlated with E2F1 activity (Figure 24B, Pearson correlation = 0.36, p value = 3.8x10-4) and tumor vs normal phosphorylation of histone H3 (Figure 24C, Pearson correlation = 0.45, p value = 7.2x10-4), which is a marker of proliferation. Finally, average RB1 phosphorylation was negatively correlated with a set of apoptosis proteins (Figure 24D, Pearson correlation = -0.29, p value = 4.6x10-3).

R=0.54, P=1.5e−08 R=0.36, P=3.8e−04 A ● B ● ●● ● ● ● ● ● ● ● 0.3 ● ● ● ● ● ●● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● 0.6 ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● 0.2 ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

●● ● 0.5 ●● ● ● ● ● ● ● ● ●

0.1 ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● E2F1 Activity

CDK2 Activity ● ● ● 0.0 0.4

● ● 0.1

− ● ● 0.3

−1 0 1 2 3 −1 0 1 2 3

R=0.45,ave_ord P=7.2e−04 R=−0.29,ave_ord P=4.6e−03 3 C ● ● D 0.15

● ●

● ● 2 ● ● ● ● ● ● ● 0.10 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● 1 ● ● ● ● ● ● ●● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● 0.05 ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

0 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● 0.00 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● 1 ● ● ● ● − ● ● ● ● ●● ● ● ● ● ●

0.05 ● ● ●

− ● ● Histone 3 phosphorylation ● ●

2 ● − Apoptosis Hallmark Enrichment ● 0.10 − −1 0 1 2 3 −1 0 1 2 3 Average RB1 Phosphorylation Average RB1 Phosphorylation

Figure 24. RB1 correlation with proliferation and apoptosis markers. Correlation of average RB1 phosphorylation change in tumor compared to normal correlation with A) ssGSEA activity scores for CDK2, B) ssGSEA enrichment of protein abundance of targets of E2F1, C) phosphorylation of histone H3.1, and D) protein abundance of apoptotic proteins.

57 Unlike tumors where RB1 is deleted or mutated, RB1 inactivation by phosphorylation is a possible drug target in colon cancer. RB1 could be re-activated by CDK inhibitors. Additionally, RB1 phosphorylation could be used as a marker of response to CDK drugs. Therefore, I examined RB1 characteristics in comparison to cell line response to CDK inhibitors. Using data from the GDSC, cell lines with higher RB1 mRNA expression were significantly more sensitive to two different CDK inhibitors (Figure 25A- B). One drug (PHA-793887) targeted CDK2/5/7, while the other (palbociclib) targeted CDK4/6. CDK2, CDK4, and CDK6 are known to phosphorylate RB1. The MCLP project assessed RB1 phosphorylation using RPPA for many of the same cell lines. RB1 phosphorylation also correlated with cell line sensitivity to these two inhibitors (Figure 25C-D).

P=0.004 P=0.002 A B ● 7.0 7.0 6.5 6.5 6.0 6.0

5.5 ● 5.5 ● pIC50 pIC50 5.0 5.0 4.5 4.5 4.0 4.0 3.5

Low RB1 mRNA High RB1 mRNA Low RB1 mRNA High RB1 mRNA

R=0.25 P=3.37e−05 R=0.35 P=3.32e−08

C ● ● D ● ● 7 ● ● ● ● ● ● 7 ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● 6 ● ● ● ● ● ●● ● ●● ● ● ● ● ●● ● ● ●● ● 6 ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ●● ● ●● ●● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ●● ● ● ● ● ● ● ● ● ● ● ●● ● ●●● ● ● ● ● ● ● ● ●●● ● ● ● ● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● pIC50 ● ● ● ● ● ● pIC50 ● ● ●● ● ● ● ● ● ● ●● ● ● ● ● ● ●

● ● 5 ●● ●● ● ●● ●● ● ●● ● ● ● 5 ● ● ● ● ●● ● ● ● ● ● ● ● ● ●● ●● ● ● ● ● ● ● ● ● ● ●● ●● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ●●● ● ● ● ● ● ● ● ●● ● ● ●● ● ● ● ● ● ● ● ●●● ● ● ● ● ● ●● ● ●● ● ● ● ●●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● 4 ● ● ● ● ● ● ● ● ● ● 4 ● ● ● ●● ● ● ● ● ● ●● ● ● ● ● ● ● ●● ●●● ● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ●

0 1 2 3 4 5 6 0 1 2 3 4 5 6 RB1 pS807/S811 RB1 pS807/S811

Figure 25. RB1 effect on cell line sensitivity to CDK inhibitors. Comparison of cell line sensitivity to A) PHA-793887 and B) palbociclib between cell lines with low RB1 mRNA expression (RSEM < 3) and high RB1 mRNA expression (RSEM ³ 6). C) Correlation of RB1 phosphorylation at pS807/S811 and response to PHA-793887 or D) palbociclib.

58 Summary of RB1 in Colon Cancer Together, in colon cancer RB1 was amplified at the gene level and this amplification persisted at the protein level (Figure 26). Furthermore, RB1 was hyperphosphorylated by active CDK2. This released E2F to drive proliferation but also contributed to inhibition of apoptosis, which gave RB1 two different ways to promote tumor progression and survival. RB1 DNA

RB1 Protein

E2F1

CDK2 E2F1 Proliferation p p Apoptosis

RB1 Phosphoprotein

Figure 26. Summary of RB1 activity in colon cancer. RB1 is amplified at the DNA level and also the proteomic level. High levels of CDK2 activity results in increased RB1 phosphorylation, releasing E2F1 to drive proliferation. Furthermore, RB1 contributes to apoptosis suppression.

Discussion Overall, phosphoproteomic data provided both a complementary and unique perspective of colon cancer compared to other data types. Phosphoproteomic data confirmed some of the colon cancer targets known by genomic or proteomic data. High phosphorylation was found on DDX21, RSL1D1, S100A11, and TOP2A, which were proteins also highly over-expressed in tumor samples compared to normal. TOP2A is involved in the cell cycle and is frequently overexpressed in cancers, including colon cancer236,237. DDX21 is an RNA that is also upregulated in cancer, correlates with proliferation, and drives rRNA processing238. RSL1D1 is involved in cellular senescence and apoptosis and has also been associated with prostate cancer severity239. Finally, S100A11 is a calcium binding protein that helps repair damaged cell membranes240. While all of these proteins are upregulated in cancer, they are not frequently mutated and are therefore not known as cancer drivers. Interestingly, although many of the proteins with highly upregulated phosphorylation sites in cancer were also upregulated at the protein level, some highly abundant sites were on proteins with low abundance in cancer. Tubulin alpha-1B chain (TUBA1B) and its associated protein MAP4 were both highly phosphorylated but downregulated at the protein level in cancer. These phosphorylation sites likely result in destabilized and promote motility in colon cancer241. Sterile a motif and HD domain containing protein 1 (SAMHD1), a nucleotide phosphatase, regulates dNTP homeostasis, has a role in genome stability, and has tumor suppressor functions242. In

59 our study, SAMHD1 was slightly downregulated 1.13x fold, but phosphorylated at T592 2.24x fold. Phosphorylation at this site might further reduce its activity by reducing protein stability242. Finally, the kinase MAP3K20, also known as ZAK, had a 3.2x fold increase of phosphorylation at S637, although the protein was downregulated 1.6x fold. ZAK can act as a cell fate switch and its downregulation in lung cancer leads to increased tumor cell survival243. However, nothing is known about the phosphorylation at S637, so it could be an interesting topic for future exploration. Phosphorylation was also high on six known cancer drivers: CDK12, DEK, PML, NPM1, RB1, and TFRC. RB1 in particular was interesting because it is a known tumor suppressor and is unusually amplified in colon cancer. The phosphorylation levels of RB1 explain the mechanism of RB1 inactivation in colon cancer. Furthermore, average RB1 phosphorylation was negatively correlated with apoptosis. RB1 has been linked to apoptosis in the literature244. Although its role is not clearly defined, hyperphosphorylated RB1 possibly binds specifically to E2F on apoptotic gene promoters while releasing E2F on cell proliferation gene promoters245. It might also control the subcellular localization of the cell survival protein Bag-1246. Finally, phosphoproteomic data showed dysregulation of several kinases in colon cancer. As expected, cell cycle regulating kinases are highly active in cancer compared to normal. However, increased phosphorylation of kinase activating sites also suggested dysregulation of a few other kinases. Some of these kinases have been linked to cancer. For example, SRPK1 expression is increased in several cancers and this promotes cancer stemness, angiogenesis, and metastasis247. PFKFB3 is also high in proliferating cells and is linked to proliferation and glycolysis248. However, these findings should be more carefully validated. MELK appeared to be a good cancer target as it was highly correlated with proliferation and small molecule inhibitors were developed as anti-cancer agents. However, complete knock-down of MELK had no effect on cell survival or proliferation rates and therefore is unlikely to have an effect on cancer patients249.

60 CHAPTER 6

CONCLUSIONS AND FUTURE DIRECTIONS

Summary Throughout this thesis, I have described the ways phosphoproteomic data can be used to identify single interesting kinases, altered signaling pathways, and ultimately pathway dysregulation at the disease level. Phosphoproteomic data provides complementary information, extends that information, and occasionally contradicts the information generated by genomic, transcriptomic, and proteomic data. Furthermore, phosphoproteomic data validates existing knowledge, generates new areas of exploration, and highlights possible therapeutic targets. This chapter summarizes the findings from this work and provides future research directions.

Evaluation of Technology While this work was clearly limited by sample size and signal resolution above noise, mass spectrometry technology continues to rapidly improve. Ten years ago, most experiments generated around 1000 phosphorylation sites250–252. Now, improvements in machine sensitivity, algorithms, and enrichment methods have expanded this 10x fold to over 10,000 quantifiable sites per experiment. Most kinases and phosphatases can already be identified using mass spectrometry and with the continued interest in this technology, the methods and analysis tools will only improve. One current extension is the enrichment and identification of kinases using kinase inhibitors attached to beads253. This allows for focused analysis on kinase signaling using small amounts of sample. Improvements in bioinformatics tools for phosphoproteomic analysis will also promote future discoveries. Tools have thus far focused on very specific areas of kinase signaling. Future work should integrate phosphatases and combine various methods. For example, tools could be developed that assess phosphorylation abundance after mutations or that allow users to visualize changes over entire signaling pathways rather than at the individual kinase level. Additionally, while many databases have collected information on kinases and phosphatases, programmatic use of this information is challenging. There are no standard terms for functions or regulation and most databases do not have downloadable data. Future work to identify phosphorylation sites on kinases that affect enzymatic activity and characterization of the regulatory mechanisms of each kinase’s activity will help improve kinase activity inference. Furthermore, new tools could integrate the various mechanisms for kinase activity inference, such as prediction from substrate phosphorylation and altered phosphorylation of regulatory sites. In my study, these two methods were complementary, and many kinases could only be identified using one of the methods. The combination might improve inference for kinases with few substrates or for kinases lacking information on their regulatory sites. Standardization is also a challenge for phosphorylation site identification. Phosphorylation sites are usually identified by their position in the protein sequence. However, protein sequences vary across databases and are frequently updated.

61 Additionally, position varies based on the splice isoform and the species. Identifying phosphorylation sites is a challenge when using multiple databases or analyzing different phosphoproteomic datasets generated using different protein sequence databases. Using the sequence surrounding a site helps, but the sequence can be identical for related proteins and slight variations can occur due to polymorphisms. Finally, the methods used here can easily be extended to study other PTMs or signaling pathways. Other PTMs are just as important as phosphorylation in both understanding cell signaling status and protein function. Additionally, many can affect the phosphorylation levels of cells. For example, a bacteria acetyltransferase can acetylate MAP2K and prevent its activating phosphorylation254. Finally, this work focused specifically on kinase signaling. However, other types, such as response to calcium or GPCR signaling are both important and frequently integrate with kinases. To understand the overall picture of cell signaling, future work could better integrate these pathways.

Kinase Signaling in Cancer One overall result from this work was the acknowledgement that kinase signaling clearly differs among subgroups of patients. These data could be used to stratify patients and understand their response to treatment. For example, patients with high levels of phosphorylated RB1 will likely respond well to CDK inhibitors, while patients with low levels of RB1 phosphorylation or protein may have little effect. Additionally, patients with kinases predicted to be highly active based on substrate phosphorylation might respond well to inhibitors of those kinases. Moreover, genomic mutations drive alterations in signaling pathways and these might be used to predict alterations to kinase signaling. Conversely, phosphorylation of kinase substrates could be used to predict the effect of understudied protein mutations. These can help predict how a patient will respond to therapy. Ideally, future work will connect kinase activity predictions from phosphoproteomic data with response to inhibitor treatment to help validate the activity predictions.

Kinase Signaling Comparison Among Tissues The addition of normal tissue greatly improves the signal-to-noise ratio. Normal tissue helps remove the general background level for an individual patient. Furthermore, stoichiometric analyses of phosphorylation are an exciting future direction. In this work, increased phosphorylation could result from increased protein abundance and/or hyper- phosphorylation of that individual site. Increasing phosphorylation regardless of protein abundance more clearly shows the actual activity of that protein. While all of these studies did have proteomic data, correcting for protein abundance is a challenge. Correction for protein abundance in the colon cancer dataset showed mostly altered abundance on cytoskeletal and stromal proteins. This might be due to tissue composition affecting normalization. Mechanisms to improve standardization, quantification, and normalization might ameliorate these effects.

62 APPENDIX A

ROC CURVE FOR SUBSTRATE PREDICTION OF PKC

1.0 0.8 0.6

MusiteDeep:0.97 Musite:0.85

0.4 GPS:0.95 iGPS:0.74 True positive rate positive True Netphorest:0.8 NetworKIN:0.74

0.2 NetPhos:0.74 PhosphoPredict:0.77 PhosphoPICK:0.7 phos_pred:0.63 0.0

0.0 0.2 0.4 0.6 0.8 1.0

False positive rate

Figure S1. ROC curve for substrate prediction of PKC. The false positive and true positive rates for PKC substrate prediction. The AUC for each tool is listed next to the tool name.

63 APPENDIX B

WEBSITE URLS FOR BIOINFORMATICS TOOLS

Name Website URL 14-3-3-Pred http://www.compbio.dundee.ac.uk/1433pred ANIA https://ania-1433.lifesci.dundee.ac.uk/prediction/webserver/index.py CEASAR http://www.phosphonetworks.org CellNOpt http://www.cellnopt.org CLUE https://cran.r-project.org/web/packages/ClueR/index.html CPhos https://hpcwebapps.cit.nih.gov/CPhos/ CrossCheck http://www.proteinguru.com/toolbox/crosscheck/ dbPAF http://dbpaf.biocuckoo.org dbPTM http://dbptm.mbc.nctu.edu.tw DEPOD http://depod.bioss.uni-freiburg.de DrugKiNET http://www.drugkinet.ca DynaPho http://140.112.52.89/dynapho/?p=100&tid=none eFIP http://research.bioinformatics.udel.edu/eFIPonline/index.php EKPD http://ekpd.biocuckoo.org GPS http://gps.biocuckoo.org GPS-Polo http://polo.biocuckoo.org HMMpTM http://aias.biol.uoa.gr/HMMpTM/ HPRD http://hprd.org HuPho http://hupho.uniroma2.it iGPS http://igps.biocuckoo.org IKAP https://github.com/marcel-mischnik/IKAP K-Map http://tanlab.ucdenver.edu/kMap KANPHOS https://kanphos.neuroinf.jp KEA2 http://www.maayanlab.net/KEA2 KIDFamMap http://gemdock.life.nctu.edu.tw/KIDFamMap/ Kin-Driver http://kin-driver.leloir.org.ar/index.php Kinannote https://sourceforge.net/projects/kinannote/ KinaseNET http://www.kinasenet.ca KinasePA http://www.maths.usyd.edu.au/u/pengyi/software/KinasePA.html KinasePhos2.0 http://kinasephos2.mbc.nctu.edu.tw KinBase http://kinase.com/web/current/ KinConform https://github.com/esbg/kinconform KinG http://king.mbu.iisc.ernet.in KinMap http://www.kinhub.org/kinmap/ KinMutBase http://structure.bmc.lu.se/idbase/KinMutBase/

64 Kinome NetworkX https://bioinfo.uth.edu/kinomenetworkX/ Kinomer http://www.compbio.dundee.ac.uk/kinomer/ KinomeSelector http://kinomeselector.jensenlab.org/index.html KinWeb http://www.itb.cnr.it/kinweb/index.htm KLIFS http://klifs.vu-compmedchem.nl KSD http://sequoia.ucsf.edu/ksd/ KSEA https://casecpb.shinyapps.io/ksea/ KSP-PUEL https://github.com/PengyiYang/KSP-PUEL MIMP http://mimp.baderlab.org MOKCa http://strubiol.icr.ac.uk/extra/mokca/ Musite http://musite.net MusiteDeep https://github.com/duolinwang/MusiteDeep NetPhorest http://netphorest.info NetPhos http://www.cbs.dtu.dk/services/NetPhos/ NetworKIN http://networkin.info/index.shtml phos_pred http://bioinformatics.ustc.edu.cn/phos_pred/ Phos3D http://phos3d.mpimp-golm.mpg.de PhoScan http://bioinfo.au.tsinghua.edu.cn/phoscan/ PhosD http://comp-sysbio.org/phosd/ PhosFox https://bitbucket.org/phintsan/phosfox PHOSIDA http://141.61.102.18/phosida/index.aspx Phosphatome http://phosphatome.net/3.0/ Phospho.ELM http://phospho.elm.eu.org/index.html Phospho3D http://www.phospho3d.org PhosphoAtlas http://cancer.ucsf.edu/phosphoatlas PhosphoLogo https://hpcwebapps.cit.nih.gov/PhosphoLogo/ PhosphoNET http://www.phosphonet.ca PhosphoNetworks http://www.phosphonetworks.org PhosphoPep http://www.phosphopep.org PhosphoPICK http://bioinf.scmb.uq.edu.au/phosphopick/phosphopick PhosphoPICK-SNP http://bioinf.scmb.uq.edu.au/phosphopick/snpanalysis PhosphoPredict http://phosphopredict.erc.monash.edu/webserver.html PhosphoSitePlus http://www.phosphosite.org PhosphoSVM http://sysbio.unl.edu/PhosphoSVM/ PhosPred-RF http://server.malab.cn/PhosPred-RF/index.jsp PhosSNP http://phossnp.biocuckoo.org PHOXTRACK http://phoxtrack.molgen.mpg.de pkaPS http://mendel.imp.ac.at/pkaPS/ PKIS http://bioinformatics.ustc.edu.cn/pkis/ PPSP http://ppsp.biocuckoo.org

65 Predikin http://predikin.biosci.uq.edu.au ProKinO http://vulcan.cs.uga.edu/prokino/about/browser ProteomeScout https://proteomescout.wustl.edu/ PSEA http://bioinfo.ncu.edu.cn/PKPred_Home.aspx PTMfunc http://ptmfunc.com PyTMs https://pymolwiki.org/index.php/Pytms RegPhos http://regphos.mbc.nctu.edu.tw/ RegPhos2.0 http://csb.cse.yzu.edu.tw/RegPhos2/ ReKINect http://rekinect.info/home RLIMS-P http://research.bioinformatics.udel.edu/rlimsp/ Scansite http://scansite4.mit.edu/#home SELPHI http://rothwebprod.mshri.on.ca:8081 Signor http://signor.uniroma2.it Sorad http://research.cs.aalto.fi/csb/software/sorad/ SubPhosDB http://bioinfo.ncu.edu.cn/SubPhosDB.aspx Swiss-Prot http://www.uniprot.org wKinMut2 http://kinmut2.bioinfo.cnio.es/KinMut2

66 APPENDIX C

BREAST CANCER SUBTYPE DRIVER SUBNETWORKS

Basal Driver Subnetwork

SHPRH

TTC28 FBXW7 BAAT

FANCF SSPO LINGO1 SZT2 GOLGA8DP

TEP1 PCSK5 RYR2 TFDP1 FANCE USP1 ALG13

CRX ARID1A KMT2C RB1 SMAD5 SPAG5 TCF20 ATN1 ACTL6B YY1AP1 RBFOX1 LRP2 FANCD2 NELL1 CRIM1 ERBB4 PTPN12 AMPD1 ASXL2 ARID1B SNTB1 CDK12 PHKA2RNF123 ZEB2CARD10 LGALS14 NBEA CROCCP2 SF3B2 NELFE PTK2B MMP11 KMT2D AFF1 STAG1 NOTCH3 RGS7 HUWE1 SNRPN MAPK14ATM FYB ATP1A1 KIF5C CNTNAP1 CSMD2 RLF TMEM57 PRX ZCCHC8 PLEKHA5 UTRNMYRIP IFT20 MTMR11 MUC12 DDR1 SNX17 PRPF40A DIAPH3 SMC1A PLEC ATXN7 KLC3 CACNA1B ABL1 VPS13A TLR3 CHEK2 CASR LAMB2 PPFIA4 UBR4 CELSR3 C1QTNF9 SNX8 PIK3CD ARHGAP9 SMAD9 ITGB4 PTPRF MRPS31 FMNL1 PREX1 BRCA1 FLNA FYN APBA1 BRIP1HSPG2MYOM3 CUL7 BRCA2 TPSAB1ADAMTS2 NCOR1 NCK1PLCG1 DLGAP4SP1 AFF2 BRWD1 CUL9 TP53 SPTAN1 PIK3R1 PRKDC KIAA2026 CDC42 SNAPIN AKAP6CACNA1F WAC WNK2 ELL2 IK TOPBP1 AMOT SHANK2 STAT3 FAM117B BATF3 CHD3 SULT1E1 ASH2L RYR1 CASK KDM6A DST SOS1 ASXL1 NCF2 SPTBN1 DROSHA PIK3CA ACTN2 CKAP5 HLF DOCK9 CMYA5 MYO18B CASP6 EHMT1 FLNC PARP1 ENO3 TTN CHAF1A PDZD2 UBC SYNE1 NFYC RABGAP1 BMX FAM208B PLIN4 LAMB1 KIF13ATULP3 CASP8 BAZ1A AKAP9CHD2 SVEP1 TNIK DISC1 NEB TNS1 NRXN1 SDCCAG3 KMT2A MYBPC2 CEP170 WNK1 ZFPM2 CARD9 PTEN DARS ASH1L H3.1 SPTB PTPRN2 SYBU MYH7 MYOM2 DYSF HDAC9 CYLD CREBBP ARID2 ZCWPW2 APC ZNF106 DMD AREL1 DYNC1H1 GREB1 GTF3C1 OBSCN ACTN3 MACF1 GATA4 CSNK1E ARHGAP21 FAM53C PLAA ANK1 PCNT RB1CC1 SETD1B COL12A1 SGCD ZNF462ACTG1 PBRM1 ARHGEF1 SPTA1 ATF7IP PDE4B EPRS TAOK1 SREBF1 TPR NUPL2 BCAR3 RANBP2 CNKSR2 BZRAP1 ESR2 XIRP2 MYH7B H4.1 LAMA3 KIF1B ADA CASZ1 EPAS1 ALMS1 ARF1 DYNC1I1 CNTNAP2NRXN2 GCC2 SHROOM3 NOMO1 SLC12A4 LAMA5 CSPG4 CDH11 MYH4 PTPRO RRP12 MYH13 RFC1 MCAM CNTNAP4 COL4A2 ANKRD11 ARID4B PAPPA2 ZFR BCLAF1 H3.2

THRAP3 RANBP10 ING1 THSD7B

TSEN2 IGSF1

HECTD1

67 Her2 Driver Subnetwork

PRPF8

HNRNPH2

HNRNPH1 ARMC8

FAM208A CWF19L1 TCF12 KMT2E CLIC5 TOM1L1 FANCM GATA3 AKR1C2 CCT3 BOD1L1 LAMTOR5 RARB TAF5 GON4L POLR1A GAPVD1 IPO4 ACTN3 PNRC1 MYO10 NME7 FAAP24 ALMS1 RECK KIF1B LPA RORA MYH8 RBL1 OBSCN COL1A1 ATP6V1B2 GPRASP1 ITGB5 GRHL3 TAF9 VAV2 COL12A1 GFAP MYBPC1 ITGAM COL6A3 TTN MYT1L TCF4 LAMA2 NEFM PRMT5 APOBEC3C RBBP6 C3 CMYA5 DYSF FN1 RABGEF1 PIAS2 MYBPC2 CTTNBP2 CHIC2 SPTA1 SPTB KLHL41 CFH SYNE2 FGR TRIM46 CLCN6 SNAPIN RBMX MAP1A POM121 HMGB1 MYH14 NEB CLU PLEKHS1 TNIK RYR1 MSH6 RANBP2 YTHDF1 YLPM1 PDE4DIP PTPN11 DOCK4 SCAMP1 CDH1JAK2 VAV3 SMG1 APPL1 DST SH2B1 NCL KRT17 NEDD9 ANK2 TNS4 BRCA1 PRKDCEHMT1 SPTBN2 ITSN2 ITK PIK3C2B TLN1 TP53 ERBB3 CRBN STAG1 PIK3CA FLNA CYP4A11 GIGYF2AMER1 FLNB GRB2 WDFY3 BLM CWF19L2 LUC7L3 ERBB2 PIK3R1 SMC3 BAI1 PRG4 APC PDGFRB B2M NUP98 DLGAP1 PLCG2AKAP6 TBC1D22B STAG2 KAT7 NCK1 TP53BP2 EXPH5 EGFR SHANK2BRCA2 PPHLN1 PTPN22 SRC COL17A1 TGDS MBD5 DOCK3 PHACTR4 LTBP4 SNX7 CRK SHANK3 PHC2 PALB2 DDX11 NUPL2 EEA1 H2.1 RASA1 ABL1 PRICKLE3 NAV1 MCM3AP NCOA6 PLK1 KIF5A PRRC2B ABL2 GNS APBB1 OGN PIK3CD NCKAP1EPHB2 TP53BP1 KAT6A HGS SLC22A3 RECQL5 ATN1 CLSTN1 FASTKSTAT2 MYH9 MAP3K10 DLC1 NBEA HSPG2 LRP2 PCDHA2PCDHA9 GLB1 DMPK SASH1 MYO18B SMARCD3 LRBA IFI16 SETDB1 ARHGAP5 NR3C1 PAK1 KAT6B STAM LTBP2 GIGYF1 LRP1

ZC3H7A NEDD4L WNK1 TOB1 ANKRA2 C5ORF42 ZFP64 SMARCA4 ZNF83 GPS2 ZBTB9 ARID1A SKIL

THSD7A RFX7 SVEP1 SCN5A SRP72 CAMSAP1 ARID1B LAPTM5 SRP68 FGF13

HERC1 DISP1

68 Luminal A Driver Subnetwork

CNTN1 SLX4 PTPRZ1 EME1

UNC45B IRS4 TLN1 STAT4GRIN2D NUP155 PLCE1 MLLT4 APBB3 JAK1 PCDHA1 RPS5 TNS3 MUS81 PIK3CG VARS2 PCDHA3MYO18B CD93 NEDD9 BEX2 PCDHA6 RPS20 FTL RGS7 CELSR2 TRIM74 WDFY3 AFF2 ATPIF1 BCR HSPD1 SHROOM2 ITIH1 ERBB2 EFS PRICKLE3 LRBA ATG16L1 HRAS BRIP1 AMBP SOS2 E2F1 DOCK3 TSKS PRKDC THOC2 INTS1 UMPS CRK HCN4 GPX2 PPFIA3 CHD6 PIK3CA MYH9 PLCG1 CUL7 DNM3 RB1CC1 HIVEP2 CHEK2 GLCE MAP4K1 ZNF281 FYN LTBP2 DEAF1 BLID CASP8 IL13RA2 PIK3R3 PIK3R1 CARNS1 ARAF RP1L1 BRCA1DDX5 VPS4B THRSP SQSTM1 MAGEC3 BEX1TNFSF13 UFD1LPSMC3IP IFI16 ACTL6B TP53 SPEN FLNBPTPN4 PDCD6IP UFD1 WNK1 ZMYM3 BTK EPSTI1 FLNC IGHM PSMC6 LPA IL24 ATM BPTF TRIP12 SH3KBP1 DNAJB6 TP53BP1 GPC3 ESR1 ADAMTS12 PTEN FBN3 GTF2I CDH1 NAT2 NOTCH2 FN1 ATP2A1 KRAS URODKMT2B APPL1 PHF8 STARD9 CELSR3 CACNA1C MTA3 CHD1 KCNQ3 CHD4 ALMS1 DSP HSPA8 HDAC1 JUP UBE2D3 GPS2 ACTN2 CDK5RAP2 RUNX1 DYSF H3.1 KDM4A CAMTA2 CALM CBFB COL7A1 KDM6A PSMA3 NCOR2 ATRX NCOA6HMGXB4NCOR1SYNE2 TBC1D4 CHGB CTNNA1 TRIM17 TTNCMYA5 CTAGE5 TSC22D1 MYBPC2 NASP CTNNB1 NEB MYH8 ACACA DKK3 APPL2 OBSCN GCC1 NAV1 KIF1A XIRP2 NR1H2 ANKRD28 PPARG SETD1B TRIP11 GATA3 CHUK RXRA ACTN4 LRRK2 YLPM1 TAF1 ANK1 RARB PCDH12 RIMBP2 EPAS1ADGRL1 WDR5MAPK8 ITGB5 LRP2 APC MAP3K1 COL4A5 SACS NEK1 SPTB HECW1 ATN1 PDE4DIP DST ARHGEF7 ZNF512B GANAB ANK2 CACNA1APOM121 KMT2C GON4L ZMYND8 VWF DCTN5 MEGF8 NUP98 MSX2 SPTA1 IKBKB PRKD2 TGS1 NINL KMT2D ZSWIM8 PCSK5 SHANK1 DLG1 BMPR2 TCEB3 LTBP4 AMER1 BZRAP1 MAPK8IP3 ZFHX3 SSPO PPP1R13B ZMYM2 NELL2 HTATSF1 RIC8A MATN2NOC2L TPR MAP2K4 RAD54B PTPN13 APLP2 JUN

TTC28 PCSK6 MAP3K4 SYNE1 SUPT5H ZFP36L1 VPS52 HCFC1 MAPK14 ARHGEF4 LMO4 CDC5L MACF1 MAPK10 BACH2 MUC12 CNTRL RAPGEF6 MAP3K5 FKBP8 WTAP CRNKL1 DCAF6 ROBO1 EFTUD2 MYT1L GRHL2 AKAP9 NRXN2 KIF26A TCF20 ERCC6L DISC1 GSE1 PRPF8 SLIT2 DMD NRXN1 OGT HLA-B ACTG1 SNX5 CEP170 GNPTAB SPTAN1 DYNC1H1 PHC3 CACNA1D PCNT PDZD2 NUP160 PSG1 ZFP36L2KIR3DL1 IFT20

AHCTF1 RAD21 STAG2 CROCCP2

SMC1A STAG1

69 Luminal B Driver Subnetwork

STIL IDH1 CEP192 CEP152 DSCAML1 TCEB3 CEP57L1 PDZD2 PLK4 MAGI3 STAG2 KRT13 CROCCP2 NRCAM NRXN1 CEP57 SMC3 SPRR2D USF3 KCNK9 HDAC9 IFT20 BLOC1S6 PLXND1NRXN2 PADI6 RALBP1 CNTNAP2 PKN1 DMD KIF3C GLTSCR1L GRIA3 PKD1 MAGED1 ASIC1 EPG5 NFYC NYNRIN DTNB LAMC1 MACF1 SFN TRAF3IP1 GRIA4 HGF ITGB4 PAPPA CDC5L MCM10 SCN2A AKAP9 PICK1 ADAMTSL4 ARHGAP31 CDH23 ERO1ARRBP1 SACS ATF7IP SYNE1 AGTPBP1 GOLGA3 ARHGEF10LABCB11 MYT1L USH1C FRYL DST PPFIA1 MCM7 RAC1 COQ9 DYNC1H1 HDAC5 KALRN DIP2A BRWD1 SERPING1 SYNE2 PRKD2 KIAA1191 MAGI1 PELI2 THBS1USP7 HECW1 ACTL6B LAMB2 NR4A1 NCOA6 VWF ZNF829 TNIK ACTN4 ATXN7 HTATSF1 CTTNBP2 DAPK1 LECT1 DRC7 RXRA PARD3 IQCB1 TNKS1BP1 MYO6 ADAR KDM2A USP1 OBSCN PINK1 CACNA1C ATM ACO1 JUN TUBG1 AGL APPL1 APLP2 LRP1 RARA CCNA2 TNKS2 FANCD2 PEG3 CACNA1A KDM3B AKAP8L TGM2 PGR TP53 L1CAM FN1 ZNF512B ALMS1 RARB APPL2 MBIP FAM120A CALM1|CALM2|CALM3 NELL2 CDC123 TTN DNAJB5 NXF1 DHX9 MYBPC2NCOR2 SOCS4 MEGF8 BRIP1 KIF1B DYSF NEB BRCA1 PCSK5 WIZ ANK2 KMT2C BCL6 NCOR1 DUSP10 KIAA1683 ILF3 RUNX1 TSC1LIMS1 NEBL XIRP2 COL6A3 CHUK CDKN1B NCL TSC22D1 ZSWIM8HMGXB4 HCFC1 CMYA5 FLNB DCTN1 TRMT2A GATA3 WDR33 DNAJB6 RANBP2 MSN RUNX1T1 EHMT2 WDR5 MAP3K1 PPARA MYLK PLCE1 ZXDC CBFB RYR1 ATN1 TLE1 MAPK14 MAP2K4 SPEN RBBP5 FOXA1 CSMD2 PIK3R1 BAZ2A MED14 FBN3 ARID4B MEPEFANCA RB1 LRP2 NCK1 PLOD3 EXTL3 PIK3R3 CPSF1 IFT140 HNRNPR NEK8 CCDC14 TEP1 GRB2 LRRK2 PTPRR USP54 ADGRL2 KMT2A DOCK3AXL PSMA3 ZFP36L1 PCDHA11 CRKPLCG1 SH3KBP1 FIP1L1 ASAP2 PIK3C2B ELK3ARHGEF11 ZMYM4 MAP2K3 SPAG5 PSIP1 MAPK4 PELI3 MUC12 SHANK2 AMBP PCDHA5 PCDHA2 PIK3CA WDFY3 IGF1RMAP4K5 FGFR2 LETM1 MYLIP DNM1 ZNF366 PCDHA8 CELSR2 MYH11 POLD1 RIT2 EYA3 WDR44EXTL1 CFH ALAS2 MYO18BANKZF1 RBBP4 GARS KCNB2 SMARCC2 CTCF SMARCA1 SERPINC1 WAS F2 PHF6 IL13RA2 ROR2 EPRS NEDD9 ARID1A RB1CC1 PTPRB NOTCH2 SORBS2

GRIN2B

70

APPENDIX D

WIKIPATHWAYS ENRICHMENT RESULTS FOR SUBTYPE DRIVER SUBNETWORKS

Table D1. WikiPathways enrichment in the basal driver subnetwork. Normalized Gene Enrichment Description Overlap Enrichment P Value FDR Set Score Score WP3651 Pathways Affected in Adenoid Cystic Carcinoma 15 3.2 4.7 2.7E-07 7.2E-05 WP244 Alpha 6 Beta 4 signaling pathway 8 1.7 4.7 1.9E-04 2.6E-02 WP314 Fas Ligand (FasL) pathway and Stress induction 9 2.4 3.7 5.1E-04 3.5E-02 of Heat Shock Proteins (HSP) regulation WP2377 Integrated Pancreatic Cancer Pathway 22 10.5 2.1 6.2E-04 3.5E-02 WP2118 Arrhythmogenic Right Ventricular 10 3.1 3.2 7.6E-04 3.5E-02 Cardiomyopathy WP383 Striated Muscle Contraction 8 2.1 3.9 7.8E-04 3.5E-02 WP707 DNA Damage Response 11 3.8 2.9 1.1E-03 4.1E-02 WP710 DNA Damage Response (only ATM dependent) 14 5.6 2.5 1.2E-03 4.1E-02

Table D2. WikiPathways enrichment in the Her2 driver subnetwork. Normalized Gene Enrichment Description Overlap Enrichment P Value FDR Set Score Score WP2261 Signaling Pathways in Glioblastoma 17 4.8 3.5 3.2E-06 0.001 WP3303 Rac1/Pak1/p38/MMP-2 pathway 14 3.8 3.7 1.4E-05 0.002 WP306 Focal Adhesion 24 9.6 2.5 2.0E-05 0.002 WP3651 Pathways Affected in Adenoid Cystic 12 3.2 3.7 5.9E-05 0.004 Carcinoma WP2377 Integrated Pancreatic Cancer Pathway 23 10.6 2.2 2.9E-04 0.013 WP2526 PDGF Pathway 9 2.3 3.9 3.1E-04 0.013 WP3680 Association Between Physico-Chemical 11 3.3 3.3 3.3E-04 0.013 Features and Toxicity Associated Pathways WP313 Signaling of Hepatocyte Growth Factor 8 2.0 3.9 7.0E-04 0.024 Receptor

Table D3. WikiPathways enrichment in the luminal A driver subnetwork. Normalized Gene Enrichment Description Overlap Enrichment P Value FDR Set Score Score WP3651 Pathways Affected in Adenoid Cystic 21 3.6 5.8 6.1E-12 1.7E-09 Carcinoma WP710 DNA Damage Response (only ATM dependent) 21 6.3 3.3 6.0E-07 8.1E-05 WP3844 PI3K-AKT-mTOR signaling pathway and 11 1.9 5.7 1.1E-06 9.8E-05 therapeutic opportunities WP2118 Arrhythmogenic Right Ventricular 14 3.5 4.0 4.2E-06 2.8E-04 Cardiomyopathy WP2377 Integrated Pancreatic Cancer Pathway 28 11.8 2.4 1.0E-05 5.5E-04 WP3915 Angiopoietin Like Protein 8 Regulatory Pathway 21 8.2 2.6 4.4E-05 1.6E-03 WP481 Insulin Signaling 24 10.1 2.4 4.7E-05 1.6E-03 WP422 MAPK Cascade 9 1.9 4.8 5.0E-05 1.6E-03 WP2261 Signaling Pathways in Glioblastoma 16 5.3 3.0 5.2E-05 1.6E-03 WP3303 Rac1/Pak1/p38/MMP-2 pathway 13 4.2 3.1 1.9E-04 5.2E-03 WP314 Fas Ligand (FasL) pathway and Stress 10 2.7 3.7 2.5E-04 6.3E-03 induction of Heat Shock Proteins (HSP) regulation WP306 Focal Adhesion 23 10.7 2.2 3.0E-04 6.4E-03

71 WP1971 Integrated Cancer Pathway 10 2.8 3.6 3.1E-04 6.4E-03 WP2034 Leptin signaling pathway 14 5.0 2.8 3.3E-04 6.4E-03 WP23 B Cell Receptor Signaling Pathway 16 6.3 2.6 3.8E-04 6.8E-03 WP673 ErbB Signaling Pathway 11 3.4 3.2 4.0E-04 6.8E-03 WP437 EGF/EGFR Signaling Pathway 22 10.4 2.1 5.3E-04 8.4E-03 WP1984 Integrated Breast Cancer Pathway 20 9.1 2.2 6.0E-04 9.1E-03 WP2526 PDGF Pathway 9 2.5 3.6 6.6E-04 9.4E-03 WP2380 Brain-Derived Neurotrophic Factor (BDNF) 20 9.3 2.2 7.3E-04 9.9E-03 signaling pathway WP383 Striated Muscle Contraction 8 2.3 3.4 1.7E-03 2.2E-02 WP615 Senescence and Autophagy in Cancer 15 6.5 2.3 1.8E-03 2.2E-02 WP1545 miRNAs involved in DNA damage response 5 1.0 5.0 2.2E-03 2.5E-02 WP195 IL-1 signaling pathway 10 3.5 2.8 2.2E-03 2.5E-02 WP2203 Thymic Stromal LymphoPoietin (TSLP) 9 3.0 3.0 2.4E-03 2.5E-02 Signaling Pathway WP712 Estrogen signaling pathway 6 1.5 4.1 2.5E-03 2.5E-02 WP2018 RANKL/RANK (Receptor activator of NFKB 10 3.6 2.8 2.5E-03 2.5E-02 (ligand)) Signaling Pathway WP3680 Association Between Physico-Chemical 10 3.7 2.7 2.9E-03 2.8E-02 Features and Toxicity Associated Pathways WP1544 MicroRNAs in cardiomyocyte hypertrophy 12 5.0 2.4 3.5E-03 3.3E-02 WP1403 AMP-activated Protein Kinase (AMPK) 10 3.8 2.6 3.8E-03 3.4E-02 Signaling WP2857 Mesodermal Commitment Pathway 15 7.1 2.1 3.9E-03 3.4E-02 WP3668 Hypothesized Pathways in Pathogenesis of 6 1.6 3.8 4.0E-03 3.4E-02 Cardiovascular Disease WP382 MAPK Signaling Pathway 20 10.7 1.9 4.4E-03 3.7E-02 WP2036 TNF related weak inducer of apoptosis 8 2.7 2.9 4.8E-03 3.7E-02 (TWEAK) Signaling Pathway WP2032 Human Thyroid Stimulating Hormone (TSH) 10 3.9 2.5 4.9E-03 3.7E-02 signaling pathway WP2870 Extracellular vesicle-mediated signaling in 6 1.7 3.6 5.0E-03 3.7E-02 recipient cells WP75 Toll-like Receptor Signaling Pathway 13 6.0 2.2 6.0E-03 4.2E-02 WP1433 Nucleotide-binding Oligomerization Domain 7 2.3 3.1 6.1E-03 4.2E-02 (NOD) pathway WP313 Signaling of Hepatocyte Growth Factor 7 2.3 3.1 6.1E-03 4.2E-02 Receptor WP2643 Nanoparticle-mediated activation of receptor 6 1.8 3.3 7.4E-03 4.9E-02 signaling WP585 Interferon type I signaling pathways 9 3.5 2.5 7.4E-03 4.9E-02

Table D4. WikiPathways enrichment in the luminal B driver subnetwork. Normalized Gene Enrichment Description Overlap Enrichment P Value FDR Set Score Score WP1544 MicroRNAs in cardiomyocyte hypertrophy 18 4.8 3.8 5.2E-07 0.0001 WP306 Focal Adhesion 25 10.1 2.5 1.6E-05 0.0022 WP2261 Signaling Pathways in Glioblastoma 16 5.1 3.2 2.8E-05 0.0025 WP437 EGF/EGFR Signaling Pathway 23 9.9 2.3 9.3E-05 0.0063 WP2911 miRNA targets in ECM and membrane 6 0.9 6.8 1.2E-04 0.0064 receptors WP2795 Cardiac Hypertrophic Response 11 3.2 3.4 2.6E-04 0.0117 WP244 Alpha 6 Beta 4 signaling pathway 8 1.8 4.4 3.1E-04 0.0119 WP2380 Brain-Derived Neurotrophic Factor (BDNF) 20 8.8 2.3 3.8E-04 0.0129 signaling pathway WP3651 Pathways Affected in Adenoid Cystic 11 3.4 3.2 4.4E-04 0.0132 Carcinoma WP2377 Integrated Pancreatic Cancer Pathway 23 11.2 2.1 6.3E-04 0.0167 WP2572 Primary Focal Segmental Glomerulosclerosis 11 3.6 3.0 7.1E-04 0.0167 FSGS

72 WP51 Regulation of Actin Cytoskeleton 18 7.9 2.3 7.4E-04 0.0167 WP2032 Human Thyroid Stimulating Hormone (TSH) 11 3.7 2.9 9.7E-04 0.0201 signaling pathway WP3844 PI3K-AKT-mTOR signaling pathway and 7 1.8 3.8 1.8E-03 0.0339 therapeutic opportunities WP2526 PDGF Pathway 8 2.4 3.3 2.1E-03 0.0367 WP2828 Bladder Cancer 7 1.9 3.7 2.2E-03 0.0367 WP1528 Physiological and Pathological Hypertrophy of 6 1.5 4.1 2.5E-03 0.0391 the Heart WP366 TGF-beta Signaling Pathway 17 8.2 2.1 2.8E-03 0.0419

73 APPENDIX E

GO BP ENRICHMENT FOR PROTEINS WITH ALTERED PHOSPHORYLATION IN COLON CANCER

Table E1. GO BP Enrichment for Proteins with Downregulated Phosphorylation Sites in Tumor. Enrichment Normalized Gene Set Description Overlap P Value FDR Score Enrichment Score GO:0006928 movement of cell or 253 181.395349 1.39474359 0 0 subcellular component GO:0007010 cytoskeleton organization 261 176.22739 1.48104106 0 0 GO:0030029 actin filament-based 179 110.594315 1.61852804 0 0 process GO:0030036 actin cytoskeleton 157 97.1576227 1.61593085 0 0 organization GO:0032970 regulation of actin filament- 87 51.1627907 1.70045455 6.44E-15 3.28E-12 based process GO:1902589 single-organism organelle 298 226.356589 1.31650685 1.18E-14 4.99E-12 organization GO:0032956 regulation of actin 74 42.8940568 1.72518072 1.40E-13 5.10E-11 cytoskeleton organization GO:0007015 actin filament organization 95 58.9147287 1.6125 4.60E-13 1.34E-10 GO:0003008 system process 161 111.627907 1.44229167 4.73E-13 1.34E-10 GO:0040011 locomotion 213 156.072351 1.36475166 7.95E-13 2.02E-10

Table E2. GO BP Enrichment for Proteins with Upregulated Phosphorylation Sites in Tumor. Enrichment Normalized Gene Set Description Overlap P Value FDR Score Enrichment Score GO:0006396 RNA processing 216 109.10422 1.97975843 0 0 GO:0006397 mRNA processing 135 71.2265289 1.89536121 0 0 GO:0008380 RNA splicing 115 60.5219638 1.90013662 0 0 GO:0016071 mRNA metabolic 157 87.6950904 1.79029407 0 0 process GO:0022613 ribonucleoprotein 89 46.5236865 1.91300404 0 0 complex biogenesis GO:0034470 ncRNA processing 71 31.7019811 2.23960767 0 0 GO:0034660 ncRNA metabolic 92 43.6416882 2.1080761 0 0 process GO:0042254 ribosome biogenesis 59 27.1731266 2.17126284 2.22E-16 7.07E-14 GO:0006364 rRNA processing 49 21.4091301 2.28874316 3.33E-16 8.48E-14 GO:0016072 rRNA metabolic process 49 21.4091301 2.28874316 3.33E-16 8.48E-14

Table E3. GO BP Enrichment for Proteins with Highly Upregulated Phosphorylation Sites in Tumor. Enrichment Normalized Gene Set Description Overlap P Value FDR Score Enrichment Score GO:0042254 ribosome biogenesis 10 1.36434109 7.32954545 4.79E-07 0.00122032 GO:0051052 regulation of DNA 11 1.88113695 5.84752747 1.25E-06 0.00158957 metabolic process GO:0022613 ribonucleoprotein 11 2.33591731 4.7090708 1.10E-05 0.00932361 complex biogenesis GO:0006364 rRNA processing 7 1.0749354 6.51201923 6.80E-05 0.03464428 GO:0016072 rRNA metabolic 7 1.0749354 6.51201923 6.80E-05 0.03464428 process GO:0006259 DNA metabolic 13 4.13436693 3.144375 0.00012511 0.04759933 process GO:0034470 ncRNA processing 8 1.59173127 5.02597403 0.00013087 0.04759933

74 APPENDIX F

COLON CANCER-ASSOCIATED PHOSPHORYLATION SITES

HGNC Symbol UniProt ID Site Median Log2(Tumor)-Log2(Normal) FDR -Log10(FDR) CLDN2 P57739 S208 1.868 2.59E-09 8.586238901 RSL1D1 O76021 S427 1.704 1.85E-14 13.73308206 RSL1D1 O76021 S443 1.6875 4.69E-12 11.32846341 MAP3K20 Q9NYL2 S637 1.6645 2.42E-13 12.6155395 PARP1 P09874 S257 1.547 1.11E-09 8.953587331 NOP2 P46087 S67 1.4895 1.00E-12 11.99799569 UBE2M P61081 S28 1.466 4.30E-12 11.36608691 DDX21 Q9NR30 S121 1.415 6.59E-15 14.18100695 NCL P19338 S67 1.4035 7.17E-15 14.14464472 FOXK1 P85037 S223 1.401 1.79E-11 10.74712642 NPM1 P06748 S260 1.384 2.31E-13 12.63704785 TOP2A P11388 S1106 1.36775 4.44E-14 13.35281059 TCOF1 Q13428 S156 1.351 1.97E-13 12.70597241 TFRC P02786 S34 1.349 1.55E-13 12.80929415 SRRM2 Q9UQ35 T2599 1.325 3.18E-10 9.496918431 RPL4 P36578 S295 1.3135 6.59E-15 14.18100695 TNS4 Q8IZW8 S350 1.293 8.32E-13 12.07968007 PNPLA2 Q96AD5 S428 1.28 1.33E-11 10.87632875 TOP2A P11388 S1247 1.2795 4.48E-12 11.34870137 FOSL2 P15408 S17 1.2595 1.86E-11 10.72961408 LIG3 P49916 S913 1.243 1.61E-14 13.79312313 RSL1D1 O76021 S392 1.219 2.51E-08 7.600505014 NUP153 P49790 S192 1.2175 6.59E-15 14.18100695 NOP2 P46087 S732 1.216 1.33E-10 9.875157307 NOLC1 Q14978 S538 1.20525 1.92E-14 13.71671823 CDK13 Q14004 S340 1.16825 2.25E-10 9.647615714 SAMHD1 Q9Y3Z3 T592 1.162 1.93E-09 8.714899114 NOC2L Q9Y3T9 S49 1.141 1.22E-14 13.91287263 ILF3 Q12906 S382 1.1355 2.04E-11 10.68932899 SUB1 P53999 S19 1.135 1.15E-11 10.9375928 RB1 P06400 T826 1.131 1.40E-12 11.853356 RPP30 P78346 S251 1.125 5.66E-13 12.24683109 PML P29590 S512 1.122 5.57E-09 8.253812672 S100A11 P31949 S6 1.122 2.99E-10 9.52434254 MAP4 P27816 S787 1.118 5.12E-08 7.291009472 CDK12 Q9NYV4 S303 1.111 4.13E-08 7.384294752 CDK2 P24941 Y15 1.101 2.17E-14 13.66453617 TMPO P42166 S184 1.098 1.96E-13 12.70724838 NUP93 Q8N1F7 S767 1.095 2.30E-12 11.6390753 NOLC1 Q14978 S563 1.0935 3.45E-10 9.462403982 MYBBP1A Q9BQG0 S1267 1.09175 6.59E-15 14.18100695 RIF1 Q5UIP0 S782 1.091 2.35E-14 13.62859382 SPP1 P10451 S234 1.088 2.61E-05 4.582709325 RB1 P06400 S807 1.082 4.87E-10 9.312207505 OSBPL3 Q9H4L5-2 S251 1.08 1.22E-09 8.913524704 MKI67 P46013 S2505 1.068 2.08E-11 10.68172081 SSFA2 P28290 S759 1.0655 3.80E-09 8.420183162 SMC3 Q9UQE7 S1065 1.06525 2.31E-13 12.63704785 DEK P35659 S306 1.06375 1.26E-12 11.89895398 TUBA1B P68363 S48 1.0445 1.83E-10 9.738036853

75 AMPD2 Q01433 S190 1.0445 1.63E-10 9.788566187 CDK13 Q14004 S342 1.043 2.12E-09 8.67438279 CDK2 P24941 T14 1.03075 1.95E-11 10.70913932 DEK P35659 S307 1.03 5.36E-14 13.27111673 NOM1 Q5C9Z4 S321 1.029 3.59E-12 11.44443685 SRSF9 Q13242 S204 1.0285 4.79E-12 11.31930057 WDR4 P57081 S391 1.02525 1.34E-12 11.8724523 NOC2L Q9Y3T9 S56 1.02225 3.70E-14 13.43162797 NOM1 Q5C9Z4 S320 1.022 2.62E-12 11.58204193 NOM1 Q5C9Z4 S317 1.0185 3.47E-12 11.45970211 NCAPG Q9BPX3 S674 1.018 2.56E-11 10.59101817 BMS1 Q14692 S552 1.014 1.44E-12 11.8427087 DEK P35659 S303 1.001 1.15E-13 12.93857546

76 REFERENCES

1. Weinstein, I. B. & Case, K. The History of Cancer Research: Introducing an AACR Centennial Series. Cancer Res 68, 6861–6862 (2008). 2. Siegel, R. L., Miller, K. D. & Jemal, A. Cancer statistics, 2017. CA: A Cancer Journal for Clinicians 67, 7–30 (2017). 3. Knudson, A. G. Two genetic hits (more or less) to cancer. Nature Reviews Cancer 1, 157–162 (2001). 4. Garber, J. E. & Offit, K. Hereditary Cancer Predisposition Syndromes. JCO 23, 276–292 (2005). 5. Bradner, J. E., Hnisz, D. & Young, R. A. Transcriptional Addiction in Cancer. Cell 168, 629–643 (2017). 6. Bahram, F., von der Lehr, N., Cetinkaya, C. & Larsson, L. G. c-Myc hot spot mutations in lymphomas result in inefficient ubiquitination and decreased proteasome-mediated turnover. Blood 95, 2104–2110 (2000). 7. Leder, A., Pattengale, P. K., Kuo, A., Stewart, T. A. & Leder, P. Consequences of widespread deregulation of the c-myc gene in transgenic mice: Multiple neoplasms and normal development. Cell 45, 485–495 (1986). 8. Sur, I. & Taipale, J. The role of enhancers in cancer. Nat. Rev. Cancer 16, 483–493 (2016). 9. Fuziwara, C. S. & Kimura, E. T. Insights into Regulation of the miR-17-92 Cluster of miRNAs in Cancer. Front Med (Lausanne) 2, (2015). 10. Bhan, A., Soleimani, M. & Mandal, S. S. Long Noncoding RNA and Cancer: A New Paradigm. Cancer Res. 77, 3965–3981 (2017). 11. Rivlin, N., Brosh, R., Oren, M. & Rotter, V. Mutations in the Tumor Suppressor Gene. Genes Cancer 2, 466–474 (2011). 12. Salesse, S. & Verfaillie, C. M. BCR/ABL: from molecular mechanisms of leukemia induction to treatment of chronic myelogenous leukemia. Oncogene 21, 8547–8559 (2002). 13. Pandolfi, P. P. Aberrant mRNA translation in cancer pathogenesis: an old concept revisited comes finally of age. Oncogene 23, 3134–3137 (2004). 14. Ashktorab, H. & Brim, H. DNA Methylation and Colorectal Cancer. Curr Colorectal Cancer Rep 10, 425–430 (2014). 15. Pinho, S. S. & Reis, C. A. Glycosylation in cancer: mechanisms and clinical implications. Nature Reviews Cancer 15, nrc3982 (2015). 16. Carvalho, S. et al. Preventing E-cadherin aberrant N-glycosylation at Asn-554 improves its critical function in gastric cancer. Oncogene 35, 1619–1631 (2016). 17. Olow, A. et al. An Atlas of the Human Kinome Reveals the Mutational Landscape Underlying Dysregulated Phosphorylation Cascades in Cancer. Cancer Res. 76, 1733–1745 (2016). 18. Radivojac, P. et al. Gain and loss of phosphorylation sites in human cancer. Bioinformatics 24, i241–i247 (2008). 19. Vogelstein, B. et al. Cancer Genome Landscapes. Science 339, 1546–1558 (2013).

77 20. Beroukhim, R. et al. The landscape of somatic copy-number alteration across human cancers. Nature 463, 899–905 (2010). 21. Tomczak, K., Czerwińska, P. & Wiznerowicz, M. The Cancer Genome Atlas (TCGA): an immeasurable source of knowledge. Contemp Oncol (Pozn) 19, A68– A77 (2015). 22. Ellis, M. J. et al. Connecting genomic alterations to cancer biology with proteomics: the NCI Clinical Proteomic Tumor Analysis Consortium. Cancer Discov 3, 1108– 1112 (2013). 23. Pereira, B. et al. The somatic mutation profiles of 2,433 breast cancers refines their genomic and transcriptomic landscapes. Nat Commun 7, 11479 (2016). 24. Consortium, T. I. C. G. International network of cancer genome projects. Nature 464, nature08987 (2010). 25. Manning, G., Whyte, D. B., Martinez, R., Hunter, T. & Sudarsanam, S. The Protein Kinase Complement of the . Science 298, 1912–1934 (2002). 26. Chalhoub, N. & Baker, S. J. PTEN and the PI3-Kinase Pathway in Cancer. Annu Rev Pathol 4, 127–150 (2009). 27. Samuels, Y. & Waldman, T. Oncogenic Mutations of PIK3CA in Human Cancers. Curr Top Microbiol Immunol 347, 21–41 (2010). 28. Karakas, B., Bachman, K. E. & Park, B. H. Mutation of the PIK3CA oncogene in human cancers. Br J Cancer 94, 455–459 (2006). 29. Burotto, M., Chiou, V. L., Lee, J.-M. & Kohn, E. C. The MAPK pathway across different malignancies: A new perspective. Cancer 120, 3446–3456 (2014). 30. Roberts, P. J. & Der, C. J. Targeting the Raf-MEK-ERK mitogen-activated protein kinase cascade for the treatment of cancer. Oncogene 26, 3291–3310 (2007). 31. Knapp, S. New opportunities for kinase drug repurposing and target discovery. British Journal of Cancer 118, 936–937 (2018). 32. Kim, A. & Cohen, M. S. The discovery of vemurafenib for the treatment of BRAF- mutated metastatic . Expert Opin Drug Discov 11, 907–916 (2016). 33. Iqbal, N. & Iqbal, N. Imatinib: A Breakthrough of Targeted Therapy in Cancer. Chemotherapy Research and Practice (2014). doi:10.1155/2014/357027 34. Bain, J. et al. The selectivity of protein kinase inhibitors: a further update. Biochem J 408, 297–315 (2007). 35. Barouch-Bentov, R. & Sauer, K. Mechanisms of Drug-Resistance in Kinases. Expert Opin Investig Drugs 20, 153–208 (2011). 36. Chrisoulidou, A. et al. Treatment compliance and severe adverse events limit the use of inhibitors in refractory thyroid cancer. Onco Targets Ther 8, 2435–2442 (2015). 37. Jensen, S. S. & Larsen, M. R. Evaluation of the impact of some experimental procedures on different phosphopeptide enrichment techniques. Rapid Commun. Mass Spectrom. 21, 3635–3645 (2007). 38. Mertins, P. et al. Ischemia in Tumors Induces Early and Sustained Phosphorylation Changes in Stress Kinase Pathways but Does Not Affect Global Protein Levels. Mol Cell Proteomics 13, 1690–1704 (2014).

78 39. Lee, D. C. H., Jones, A. R. & Hubbard, S. J. Computational phosphoproteomics: From identification to localization. Proteomics 15, 950–963 (2015). 40. Chalkley, R. J. & Clauser, K. R. Modification Site Localization Scoring: Strategies and Performance. Mol Cell Proteomics 11, 3–14 (2012). 41. Adam, K. & Hunter, T. Histidine kinases and the missing phosphoproteome from prokaryotes to eukaryotes. Lab. Invest. (2017). doi:10.1038/labinvest.2017.118 42. Park, J. et al. Building a human kinase gene repository: Bioinformatics, molecular cloning, and functional validation. Proc Natl Acad Sci U S A 102, 8114–8119 (2005). 43. Chen, M. J., Dixon, J. E. & Manning, G. Genomics and evolution of protein phosphatases. Sci Signal 10, (2017). 44. Hernandez-Armenta, C., Ochoa, D., Gonçalves, E., Saez-Rodriguez, J. & Beltrao, P. Benchmarking substrate-based kinase activity inference using phosphoproteomic data. Bioinformatics 33, 1845–1851 (2017). 45. Wiredja, D. D., Koyutürk, M. & Chance, M. R. The KSEA App: a web-based tool for kinase activity inference from quantitative phosphoproteomics. Bioinformatics (2017). doi:10.1093/bioinformatics/btx415 46. Henry, V. J., Bandrowski, A. E., Pepin, A.-S., Gonzalez, B. J. & Desfeux, A. OMICtools: an informative directory for multi-omic data analysis. Database (Oxford) 2014, (2014). 47. The UniProt Consortium. UniProt: the universal protein knowledgebase. Nucleic Acids Res. 45, D158–D169 (2017). 48. Duan, G., Li, X. & Köhn, M. The human DEPhOsphorylation database DEPOD: a 2015 update. Nucleic Acids Res. 43, D531-535 (2015). 49. Li, T., Li, F. & Zhang, X. Prediction of kinase-specific phosphorylation sites with sequence features by a log-odds ratio approach. Proteins 70, 404–414 (2008). 50. Fan, W. et al. Prediction of protein kinase-specific phosphorylation sites in hierarchical structure using functional information and random forest. Amino Acids 46, 1069–1078 (2014). 51. Miller, M. L. et al. Linear motif atlas for phosphorylation-dependent signaling. Sci Signal 1, ra2 (2008). 52. Linding, R. et al. Systematic discovery of in vivo phosphorylation networks. Cell 129, 1415–1426 (2007). 53. Song, C. et al. Systematic analysis of protein phosphorylation networks from phosphoproteomic data. Mol. Cell Proteomics 11, 1070–1083 (2012). 54. Xue, Y. et al. GPS 2.1: enhanced prediction of kinase-specific phosphorylation sites with an algorithm of motif length selection. Protein Eng. Des. Sel. 24, 255– 260 (2011). 55. Song, J. et al. PhosphoPredict: A bioinformatics tool for prediction of human kinase-specific phosphorylation substrates and sites by integrating heterogeneous feature selection. Scientific Reports 7, 6862 (2017). 56. Wang, D. et al. MusiteDeep: a deep-learning framework for general and kinase- specific phosphorylation site prediction. Bioinformatics 33, 3909–3916 (2017).

79 57. Patrick, R., Lê Cao, K.-A., Kobe, B. & Bodén, M. PhosphoPICK: modelling cellular context to map kinase-substrate phosphorylation events. Bioinformatics 31, 382– 389 (2015). 58. Blom, N., Gammeltoft, S. & Brunak, S. Sequence and structure-based prediction of eukaryotic protein phosphorylation sites. J. Mol. Biol. 294, 1351–1362 (1999). 59. Gao, J., Thelen, J. J., Dunker, A. K. & Xu, D. Musite, a Tool for Global Prediction of General and Kinase-specific Phosphorylation Sites. Molecular & Cellular Proteomics : MCP 9, 2586 (2010). 60. Neuberger, G., Schneider, G. & Eisenhaber, F. pkaPS: prediction of phosphorylation sites with the simplified kinase-substrate binding model. Biol. Direct 2, 1 (2007). 61. Huang, K.-Y. et al. dbPTM 2016: 10-year anniversary of a resource for post- translational modification of proteins. Nucleic Acids Res. 44, D435-446 (2016). 62. Sing, T., Sander, O., Beerenwinkel, N. & Lengauer, T. ROCR: visualizing classifier performance in R. Bioinformatics 21, 3940–3941 (2005). 63. Wilkes, E. H., Casado, P., Rajeeve, V. & Cutillas, P. R. Kinase activity ranking using phosphoproteomics data (KARP) quantifies the contribution of protein kinases to the regulation of cell viability. Mol. Cell Proteomics 16, 1694–1704 (2017). 64. Lachmann, A. & Ma’ayan, A. KEA: kinase enrichment analysis. Bioinformatics 25, 684–686 (2009). 65. Weidner, C., Fischer, C. & Sauer, S. PHOXTRACK-a tool for interpreting comprehensive datasets of post-translational modifications of proteins. Bioinformatics 30, 3410–3411 (2014). 66. Hornbeck, P. V. et al. PhosphoSitePlus, 2014: mutations, PTMs and recalibrations. Nucleic Acids Res. 43, D512-520 (2015). 67. Perfetto, L. et al. SIGNOR: a database of causal relationships between biological entities. Nucleic Acids Res 44, D548–D554 (2016). 68. Mischnik, M. et al. IKAP: A heuristic framework for inference of kinase activities from Phosphoproteomics data. Bioinformatics 32, 424–431 (2016). 69. Wang, J. WebGestaltR: The R Version of WebGestalt. (2017). 70. Keshava Prasad, T. S. et al. Human Protein Reference Database--2009 update. Nucleic Acids Res. 37, D767-772 (2009). 71. Dinkel, H. et al. Phospho.ELM: a database of phosphorylation sites--update 2011. Nucleic Acids Res. 39, D261-267 (2011). 72. Wang, Y. et al. EKPD: a hierarchical database of eukaryotic protein kinases and protein phosphatases. Nucleic Acids Res. 42, D496-502 (2014). 73. Martin, D. M. A., Miranda-Saavedra, D. & Barton, G. J. Kinomer v. 1.0: a database of systematically classified eukaryotic protein kinases. Nucleic Acids Res. 37, D244-250 (2009). 74. Buzko, O. & Shokat, K. M. A kinase sequence database: sequence alignments and family assignment. Bioinformatics 18, 1274–1275 (2002). 75. Krupa, A., Abhinandan, K. R. & Srinivasan, N. KinG: a database of protein kinases in genomes. Nucleic Acids Res. 32, D153-155 (2004).

80 76. Ortutay, C., Väliaho, J., Stenberg, K. & Vihinen, M. KinMutBase: a registry of disease-causing mutations in protein kinase domains. Hum. Mutat. 25, 435–442 (2005). 77. Milanesi, L. et al. Systematic analysis of human kinase genes: a large number of genes and events result in functional and structural diversity. BMC Bioinformatics 6, S20 (2005). 78. McSkimming, D. I. et al. ProKinO: A Unified Resource for Mining the Cancer Kinome. Human Mutation 36, 175 (2015). 79. Richardson, C. J. et al. MoKCa database--mutations of kinases in cancer. Nucleic Acids Res. 37, D824-831 (2009). 80. Simonetti, F. L., Tornador, C., Nabau-Moretó, N., Molina-Vila, M. A. & Marino- Buslje, C. Kin-Driver: a database of driver mutations in protein kinases. Database (Oxford) 2014, (2014). 81. van Linden, O. P. J., Kooistra, A. J., Leurs, R., de Esch, I. J. P. & de Graaf, C. KLIFS: a knowledge-based structural database to navigate kinase-ligand interaction space. J. Med. Chem. 57, 249–277 (2014). 82. Liberti, S. et al. HuPho: the human phosphatase portal. FEBS J. 280, 379–387 (2013). 83. Kooistra, A. J. et al. KLIFS: a structural kinase-ligand interaction database. Nucleic Acids Res. 44, D365-371 (2016). 84. Bodenmiller, B. et al. PhosphoPep--a phosphoproteome resource for systems biology research in Drosophila Kc167 cells. Mol. Syst. Biol. 3, 139 (2007). 85. Bodenmiller, B. et al. PhosphoPep--a database of protein phosphorylation sites in model organisms. Nat. Biotechnol. 26, 1339–1340 (2008). 86. Diella, F. et al. Phospho.ELM: a database of experimentally verified phosphorylation sites in eukaryotic proteins. BMC Bioinformatics 5, 79 (2004). 87. Diella, F., Gould, C. M., Chica, C., Via, A. & Gibson, T. J. Phospho.ELM: a database of phosphorylation sites--update 2008. Nucleic Acids Res. 36, D240-244 (2008). 88. Zanzoni, A. et al. Phospho3D 2.0: an enhanced database of three-dimensional structures of phosphorylation sites. Nucleic Acids Res. 39, D268-271 (2011). 89. Peri, S. et al. Development of human protein reference database as an initial platform for approaching systems biology in humans. Genome Res. 13, 2363–2371 (2003). 90. Mishra, G. R. et al. Human protein reference database--2006 update. Nucleic Acids Res. 34, D411-414 (2006). 91. Gnad, F., Gunawardena, J. & Mann, M. PHOSIDA 2011: the posttranslational modification database. Nucleic Acids Res. 39, D253-260 (2011). 92. Gnad, F. et al. PHOSIDA (phosphorylation site database): management, structural and evolutionary investigation, and prediction of phosphosites. Genome Biol. 8, R250 (2007). 93. Beltrao, P. et al. Systematic functional prioritization of protein posttranslational modifications. Cell 150, 413–425 (2012).

81 94. Chen, X., Shi, S.-P., Suo, S.-B., Xu, H.-D. & Qiu, J.-D. Proteomic analysis and prediction of human phosphorylation sites in subcellular level reveal subcellular specificity. Bioinformatics 31, 194–200 (2015). 95. Huang, K.-Y. et al. RegPhos 2.0: an updated resource to explore protein kinase- substrate phosphorylation networks in mammals. Database (Oxford) 2014, bau034 (2014). 96. Lee, T.-Y., Bo-Kai Hsu, J., Chang, W.-C. & Huang, H.-D. RegPhos: a system to explore the protein kinase-substrate phosphorylation network in humans. Nucleic Acids Res. 39, D777-787 (2011). 97. Tinti, M. et al. ANIA: ANnotation and Integrated Analysis of the 14-3-3 interactome. Database (Oxford) 2014, bat085 (2014). 98. Tinti, M., Johnson, C., Toth, R., Ferrier, D. E. K. & Mackintosh, C. Evolution of signal multiplexing by 14-3-3-binding 2R-ohnologue protein families in the vertebrates. Open Biol 2, 120103 (2012). 99. Hu, J. et al. PhosphoNetworks: a database for human phosphorylation networks. Bioinformatics 30, 141–142 (2014). 100. Cheng, F., Jia, P., Wang, Q. & Zhao, Z. Quantitative network mapping of the human kinome interactome reveals new clues for rational kinase inhibitor discovery and individualized cancer therapy. Oncotarget 5, 3697–3710 (2014). 101. Matlock, M. K., Holehouse, A. S. & Naegle, K. M. ProteomeScout: a repository and analysis resource for post-translational modifications and proteins. Nucleic Acids Res. 43, D521-530 (2015). 102. Naegle, K. M., Welsch, R. E., Yaffe, M. B., White, F. M. & Lauffenburger, D. A. MCAM: multiple clustering analysis methodology for deriving hypotheses and insights from high-throughput proteomic datasets. PLoS Comput. Biol. 7, e1002119 (2011). 103. Lee, T.-Y. et al. dbPTM: an information repository of protein post-translational modification. Nucleic Acids Res. 34, D622-627 (2006). 104. Lu, C.-T. et al. DbPTM 3.0: an informative resource for investigating substrate site specificity and functional association of protein post-translational modifications. Nucleic Acids Res. 41, D295-305 (2013). 105. Ullah, S. et al. dbPAF: an integrative database of protein phosphorylation in animals and fungi. Scientific Reports 6, srep23534 (2016). 106. Nagai, T., Yoshimoto, J., Kannon, T., Kuroda, K. & Kaibuchi, K. Phosphorylation Signals in Striatal Medium Spiny Neurons. Trends Pharmacol. Sci. 37, 858–871 (2016). 107. Safaei, J., Maňuch, J., Gupta, A., Stacho, L. & Pelech, S. Prediction of 492 human protein kinase substrate specificities. Proteome Sci 9 Suppl 1, S6 (2011). 108. Quintaje, S. B. & Orchard, S. The Annotation of Both Human and Mouse in UniProtKB/Swiss-Prot: One Small Step in Manual Annotation, One Giant Leap for Full Comprehension of Genomes*. Molecular & Cellular Proteomics : MCP 7, 1409 (2008).

82 109. Lo Surdo, P., Calderone, A., Cesareni, G. & Perfetto, L. SIGNOR: A Database of Causal Relationships Between Biological Entities-A Short Guide to Searching and Browsing. Curr Protoc Bioinformatics 58, 8.23.1-8.23.16 (2017). 110. Gong, W. et al. PepCyber:P~PEP: a database of human protein protein interactions mediated by phosphoprotein-binding domains. Nucleic Acids Res. 36, D679-683 (2008). 111. Davezac, N., Baldin, V., Blot, J., Ducommun, B. & Tassan, J.-P. Human pEg3 kinase associates with and phosphorylates CDC25B phosphatase: a potential role for pEg3 in cell cycle regulation. Oncogene 21, 7630–7641 (2002). 112. Chauhan, D. et al. SHP2 mediates the protective effect of interleukin-6 against dexamethasone-induced apoptosis in multiple myeloma cells. J. Biol. Chem. 275, 27845–27850 (2000). 113. Trost, B. & Kusalik, A. Computational prediction of eukaryotic phosphorylation sites. Bioinformatics 27, 2927–2935 (2011). 114. Miller, M. L. & Blom, N. Kinase-specific prediction of protein phosphorylation sites. Methods Mol. Biol. 527, 299–310, x (2009). 115. Wong, Y.-H. et al. KinasePhos 2.0: a web server for identifying protein kinase- specific phosphorylation sites based on sequences and coupling patterns. Nucleic Acids Res. 35, W588-594 (2007). 116. Xue, Y., Li, A., Wang, L., Feng, H. & Yao, X. PPSP: prediction of PK-specific phosphorylation site with Bayesian decision theory. BMC Bioinformatics 7, 163 (2006). 117. Saunders, N. F. W., Brinkworth, R. I., Huber, T., Kemp, B. E. & Kobe, B. Predikin and PredikinDB: a computational framework for the prediction of protein kinase peptide specificity and an associated database of phosphorylation sites. BMC Bioinformatics 9, 245 (2008). 118. Iakoucheva, L. M. et al. The importance of intrinsic disorder for protein phosphorylation. Nucleic Acids Res. 32, 1037–1049 (2004). 119. Durek, P., Schudoma, C., Weckwerth, W., Selbig, J. & Walther, D. Detection and characterization of 3D-signature phosphorylation site motifs and their contribution towards improved phosphorylation site prediction in proteins. BMC Bioinformatics 10, 117 (2009). 120. Newman, R. H. et al. Construction of human activity-based phosphorylation networks. Molecular Systems Biology 9, 655 (2013). 121. Horn, H. et al. KinomeXplorer: an integrated platform for kinome biology studies. Nat. Methods 11, 603–604 (2014). 122. Liu, Z. et al. Systematic analysis of the Plk-mediated phosphoregulation in eukaryotes. Brief Bioinform 14, 344–360 (2013). 123. Tsaousis, G. N., Bagos, P. G. & Hamodrakas, S. J. HMMpTM: improving transmembrane protein topology prediction using phosphorylation and glycosylation site prediction. Biochim. Biophys. Acta 1844, 316–322 (2014). 124. Zou, L. et al. PKIS: computational identification of protein kinases for experimentally discovered protein phosphorylation sites. BMC Bioinformatics 14, 247 (2013).

83 125. Dou, Y., Yao, B. & Zhang, C. PhosphoSVM: prediction of phosphorylation sites by integrating various protein sequence attributes with a support vector machine. Amino Acids 46, 1459–1469 (2014). 126. Suo, S.-B., Qiu, J.-D., Shi, S.-P., Chen, X. & Liang, R.-P. PSEA: Kinase-specific prediction and analysis of human phosphorylation substrates. Sci Rep 4, 4524 (2014). 127. Obenauer, J. C., Cantley, L. C. & Yaffe, M. B. Scansite 2.0: Proteome-wide prediction of cell signaling interactions using short sequence motifs. Nucleic Acids Res. 31, 3635–3641 (2003). 128. Yang, P., Humphrey, S. J., James, D. E., Yang, Y. H. & Jothi, R. Positive-unlabeled ensemble learning for kinase substrate prediction from dynamic phosphoproteomics data. Bioinformatics 32, 252–259 (2016). 129. Qin, G.-M., Li, R.-Y. & Zhao, X.-M. PhosD: inferring kinase-substrate interactions based on protein domains. Bioinformatics 33, 1197–1204 (2017). 130. Blom, N., Sicheritz-Pontén, T., Gupta, R., Gammeltoft, S. & Brunak, S. Prediction of post-translational glycosylation and phosphorylation of proteins from the amino acid sequence. Proteomics 4, 1633–1649 (2004). 131. Wei, L., Xing, P., Tang, J. & Zou, Q. PhosPred-RF: A Novel Sequence-Based Predictor for Phosphorylation Sites Using Sequential Information Only. IEEE Trans Nanobioscience 16, 240–247 (2017). 132. Terfve, C. et al. CellNOptR: a flexible toolkit to train protein signaling networks to data using multiple logic formalisms. BMC Syst Biol 6, 133 (2012). 133. Äijö, T., Granberg, K. & Lähdesmäki, H. Sorad: a systems biology approach to predict and modulate dynamic signaling pathway response from phosphoproteome time-course measurements. Bioinformatics 29, 1283–1291 (2013). 134. Söderholm, S., Hintsanen, P., Öhman, T., Aittokallio, T. & Nyman, T. A. PhosFox: a bioinformatics tool for peptide-level processing of LC-MS/MS-based phosphoproteomic data. Proteome Sci 12, 36 (2014). 135. Petsalaki, E. et al. SELPHI: correlation-based identification of kinase-associated networks from global phospho-proteomics data sets. Nucleic Acids Res. 43, W276- 282 (2015). 136. Hsu, C.-L., Wang, J.-K., Lu, P.-C., Huang, H.-C. & Juan, H.-F. DynaPho: a web platform for inferring the dynamics of time-series phosphoproteomics. Bioinformatics (2017). doi:10.1093/bioinformatics/btx443 137. Yang, P. et al. KinasePA: Phosphoproteomics data annotation using hypothesis driven kinase perturbation analysis. Proteomics 16, 1868–1871 (2016). 138. Ren, J. et al. PhosSNP for Systematic Analysis of Genetic Polymorphisms That Influence Protein Phosphorylation. Molecular & Cellular Proteomics : MCP 9, 623 (2010). 139. Wagih, O., Reimand, J. & Bader, G. D. MIMP: predicting the impact of mutations on kinase-substrate phosphorylation. Nat. Methods 12, 531–533 (2015). 140. Creixell, P. et al. Kinome-wide Decoding of Network-Attacking Mutations Rewiring Cancer Signaling. Cell 163, 202–217 (2015).

84 141. Patrick, R., Kobe, B., Lê Cao, K.-A. & Bodén, M. PhosphoPICK-SNP: quantifying the effect of amino acid variants on protein phosphorylation. Bioinformatics 33, 1773–1781 (2017). 142. Vazquez, M., Pons, T., Brunak, S., Valencia, A. & Izarzugaza, J. M. G. wKinMut-2: Identification and Interpretation of Pathogenic Variants in Human Protein Kinases. Hum. Mutat. 37, 36–42 (2016). 143. Chiu, Y.-Y. et al. KIDFamMap: a database of kinase-inhibitor-disease family maps for kinase inhibitor selectivity and binding mechanisms. Nucleic Acids Res. 41, D430-440 (2013). 144. Kim, J., Yoo, M., Kang, J. & Tan, A. C. K-Map: connecting kinases with therapeutics for drug repurposing and development. Hum. Genomics 7, 20 (2013). 145. Eid, S., Turk, S., Volkamer, A., Rippmann, F. & Fulle, S. KinMap: a web-based tool for interactive navigation through human kinome data. BMC Bioinformatics 18, 16 (2017). 146. Warnecke, A., Sandalova, T., Achour, A. & Harris, R. A. PyTMs: a useful PyMOL plugin for modeling common post-translational modifications. BMC Bioinformatics 15, 370 (2014). 147. Douglass, J. et al. Identifying protein kinase target preferences using mass spectrometry. Am. J. Physiol., Cell Physiol. 303, C715-727 (2012). 148. Torii, M. et al. RLIMS-P: an online text-mining tool for literature-based extraction of protein phosphorylation information. Database (Oxford) 2014, (2014). 149. Arighi, C. N. et al. eFIP: a tool for mining functional impact of phosphorylation from literature. Methods Mol. Biol. 694, 63–75 (2011). 150. Zhao, B., Pisitkun, T., Hoffert, J. D., Knepper, M. A. & Saeed, F. CPhos: a program to calculate and visualize evolutionarily conserved functional phosphorylation sites. Proteomics 12, 3299–3303 (2012). 151. Madeira, F. et al. 14-3-3-Pred: improved methods to predict 14-3-3-binding phosphopeptides. Bioinformatics 31, 2276–2283 (2015). 152. McSkimming, D. I., Rasheed, K. & Kannan, N. Classifying kinase conformations using a machine learning approach. BMC Bioinformatics 18, 86 (2017). 153. Goldberg, J. M. et al. Kinannote, a computer program to identify and classify members of the eukaryotic protein kinase superfamily. Bioinformatics 29, 2387– 2394 (2013). 154. Najafov, J. & Najafov, A. CrossCheck: an open-source web tool for high-throughput screen data analysis. Sci Rep 7, 5855 (2017). 155. Patra, K. C. et al. Hexokinase 2 is required for tumor initiation and maintenance and its systemic deletion is therapeutic in mouse models of cancer. Cancer Cell 24, 213–228 (2013). 156. Gao, Q. et al. Driver Fusions and Their Implications in the Development and Treatment of Human Cancers. Cell Rep 23, 227-238.e3 (2018). 157. Ding, L. et al. Perspective on Oncogenic Processes at the End of the Beginning of Cancer Genomics. Cell 173, 305-320.e10 (2018). 158. Mertins, P. et al. Proteogenomics connects somatic mutations to signalling in breast cancer. Nature 534, 55–62 (2016).

85 159. Zhang, B. et al. Proteogenomic characterization of human colon and rectal cancer. Nature 513, 382–387 (2014). 160. Giansanti, P., Tsiatsiani, L., Low, T. Y. & Heck, A. J. R. Six alternative proteases for mass spectrometry–based proteomics beyond trypsin. Nature Protocols 11, 993–1006 (2016). 161. Fiehn, O. & Weckwerth, W. Mass Spectrometry: Quantitation. in Encyclopedic Reference of Genomics and Proteomics in Molecular Medicine 1030–1034 (Springer, Berlin, Heidelberg, 2006). doi:10.1007/3-540-29623-9_3550 162. Cooper, B., Feng, J. & Garrett, W. M. Relative, label-free protein quantitation: spectral counting error statistics from nine replicate MudPIT samples. J. Am. Soc. Mass Spectrom. 21, 1534–1546 (2010). 163. Dowle, A. A., Wilson, J. & Thomas, J. R. Comparing the Diagnostic Classification Accuracy of iTRAQ, Peak-Area, Spectral-Counting, and emPAI Methods for Relative Quantification in Expression Proteomics. J. Proteome Res. 15, 3550–3562 (2016). 164. Latosinska, A. et al. Comparative Analysis of Label-Free and 8-Plex iTRAQ Approach for Quantitative Tissue Proteomic Analysis. PLOS ONE 10, e0137048 (2015). 165. Yang, D.-S. et al. Design and synthesis of an immobilized metal affinity chromatography and metal oxide affinity chromatography hybrid material for improved phosphopeptide enrichment. Journal of Chromatography A 1505, 56–62 (2017). 166. Daub, H. et al. Kinase-Selective Enrichment Enables Quantitative Phosphoproteomics of the Kinome across the Cell Cycle. Molecular Cell 31, 438– 448 (2008). 167. Zhang, H. et al. Integrated Proteogenomic Characterization of Human High-Grade Serous Ovarian Cancer. Cell 166, 755–765 (2016). 168. Cerami, E. et al. The cBio Cancer Genomics Portal: An Open Platform for Exploring Multidimensional Cancer Genomics Data. Cancer Discov 2, 401–404 (2012). 169. Gao, J. et al. Integrative analysis of complex cancer genomics and clinical profiles using the cBioPortal. Sci Signal 6, pl1 (2013). 170. Uhlén, M. et al. Proteomics. Tissue-based map of the human proteome. Science 347, 1260419 (2015). 171. Barbie, D. A. et al. Systematic RNA interference reveals that oncogenic KRAS- driven cancers require TBK1. Nature 462, 108–112 (2009). 172. Hänzelmann, S., Castelo, R. & Guinney, J. GSVA: gene set variation analysis for microarray and RNA-seq data. BMC Bioinformatics 14, 7 (2013). 173. Chen, H. & Boutros, P. C. VennDiagram: a package for the generation of highly- customizable Venn and Euler diagrams in R. BMC Bioinformatics 12, 35 (2011). 174. Kolde, R. pheatmap: Pretty Heatmaps. R package 1.0.10, (2018). 175. Takaku, M., Grimm, S. A. & Wade, P. A. GATA3 in breast cancer: tumor suppressor or oncogene? Gene Expr 16, 163–168 (2015).

86 176. Li, N.-S. et al. LKB1/AMPK inhibits TGF-β1 production and the TGF-β signaling pathway in breast cancer cells. Tumour Biol. 37, 8249–8258 (2016). 177. Sengupta, S. et al. Activation of tumor suppressor LKB1 by honokiol abrogates cancer stem-like phenotype in breast cancer via inhibition of oncogenic Stat3. Oncogene 36, 5709–5721 (2017). 178. Chen, I.-C. et al. Clinical Relevance of Liver Kinase B1(LKB1) Protein and Gene Expression in Breast Cancer. Sci Rep 6, 21374 (2016). 179. Chiker, S. et al. Cdk5 promotes DNA replication stress checkpoint activation through RPA-32 phosphorylation, and impacts on metastasis free survival in breast cancer patients. Cell Cycle 14, 3066–3078 (2015). 180. Pozo, K. & Bibb, J. A. The Emerging Role of Cdk5 in Cancer. Trends Cancer 2, 606–618 (2016). 181. Rouillard, A. D. et al. The harmonizome: a collection of processed datasets gathered to serve and mine knowledge about genes and proteins. Database (Oxford) 2016, (2016). 182. Kurek, D., Garinis, G. A., van Doorninck, J. H., van der Wees, J. & Grosveld, F. G. Transcriptome and phenotypic analysis reveals Gata3-dependent signalling pathways in murine hair follicles. Development 134, 261–272 (2007). 183. Galmarini, C. M. et al. 5’-(3’)- mRNA levels in blast cells are a prognostic factor in patients treated with cytarabine. Haematologica 89, 617–619 (2004). 184. Parker, J. S. et al. Supervised risk predictor of breast cancer based on intrinsic subtypes. J. Clin. Oncol. 27, 1160–1167 (2009). 185. Sørlie, T. et al. Gene expression patterns of breast carcinomas distinguish tumor subclasses with clinical implications. Proc Natl Acad Sci U S A 98, 10869–10874 (2001). 186. Fallahpour, S., Navaneelan, T., De, P. & Borgo, A. Breast cancer survival by molecular subtype: a population-based analysis of cancer registry data. CMAJ Open 5, E734–E739 (2017). 187. Mao, J.-H., Diest, P. J. van, Perez-Losada, J. & Snijders, A. M. Revisiting the impact of age and molecular subtype on overall survival after radiotherapy in breast cancer patients. Scientific Reports 7, 12587 (2017). 188. Hofree, M., Shen, J. P., Carter, H., Gross, A. & Ideker, T. Network-based stratification of tumor mutations. Nat. Methods 10, 1108–1115 (2013). 189. Zhu, J. et al. Deciphering genomic alterations in colorectal cancer through transcriptional subtype-based network analysis. PLoS ONE 8, e79282 (2013). 190. Zhang, B. et al. Relating protein adduction to gene expression changes: a systems approach. Mol Biosyst 7, 2118–2127 (2011). 191. Das, J. & Yu, H. HINT: High-quality protein interactomes and their applications in understanding human disease. BMC Syst Biol 6, 92 (2012). 192. Csardi, G. & Nepusz, T. The igraph software package for complex network research. InterJournal Complex Systems, 1695 (2006). 193. Slenter, D. N. et al. WikiPathways: a multifaceted pathway database bridging metabolomics to other omics research. Nucleic Acids Res 46, D661–D667 (2018).

87 194. Shannon, P. et al. Cytoscape: a software environment for integrated models of biomolecular interaction networks. Genome Res. 13, 2498–2504 (2003). 195. Blaydes, J. P. et al. Stoichiometric Phosphorylation of Human p53 at Ser315Stimulates p53-dependent Transcription. J. Biol. Chem. 276, 4699–4708 (2001). 196. Gabant, G. et al. Autophosphorylated residues involved in the regulation of human chk2 in vitro. J. Mol. Biol. 380, 489–503 (2008). 197. Chen, B. P. C. et al. Ataxia Telangiectasia Mutated (ATM) Is Essential for DNA- PKcs Phosphorylations at the Thr-2609 Cluster upon DNA Double Strand Break. Journal of Biological Chemistry 282, 6582–6587 (2007). 198. Zhang, M. H., Man, H. T., Zhao, X. D., Dong, N. & Ma, S. L. Estrogen receptor- positive breast cancer molecular signatures and therapeutic potentials (Review). Biomed Rep 2, 41–52 (2014). 199. Paplomata, E. & O’Regan, R. The PI3K/AKT/mTOR pathway in breast cancer: targets, trials and biomarkers. Ther Adv Med Oncol 6, 154–166 (2014). 200. Paulson, A. K. et al. MET and ERBB2 are coexpressed in ERBB2+ breast cancer and contribute to innate resistance. Mol. Cancer Res. 11, 1112–1121 (2013). 201. Zagouri, F. et al. Low Protein Expression of MET in ER-positive and HER2-positive Breast Cancer. Anticancer Res 34, 1227–1231 (2014). 202. Bertucci, F., Finetti, P. & Birnbaum, D. Basal Breast Cancer: A Complex and Deadly Molecular Subtype. Curr Mol Med 12, 96–110 (2012). 203. Ribeiro, E. et al. Triple Negative Breast Cancers Have a Reduced Expression of DNA Repair Genes. PLOS ONE 8, e66243 (2013). 204. Perou, C. M. et al. Molecular portraits of human breast tumours. Nature 406, 747– 752 (2000). 205. American Cancer Society. Cancer Facts & Figures 2018. (American Cancer Society, 2018). 206. Dienstmann, R., Salazar, R. & Tabernero, J. Personalizing colon cancer adjuvant therapy: selecting optimal treatments for individual patients. J. Clin. Oncol. 33, 1787–1796 (2015). 207. Comprehensive Molecular Characterization of Human Colon and Rectal Cancer. Nature 487, 330–337 (2012). 208. Gonsalves, W. I. et al. Patient and tumor characteristics and BRAF and KRAS mutations in colon cancer, NCCTG/Alliance N0147. J. Natl. Cancer Inst. 106, (2014). 209. Oikonomou, E., Koustas, E., Goulielmaki, M. & Pintzas, A. BRAF vs RAS oncogenes: are mutations of the same pathway equal? differential signalling and therapeutic implications. Oncotarget 5, 11752–11777 (2014). 210. Chen, J., Elfiky, A., Han, M., Chen, C. & Saif, M. W. The Role of Src in Colon Cancer and Its Therapeutic Implications. Clinical Colorectal Cancer 13, 5–13 (2014). 211. Fruman, D. A. & Rommel, C. PI3K and Cancer: Lessons, Challenges and Opportunities. Nat Rev Drug Discov 13, 140–156 (2014).

88 212. Dyson, N. J. RB1: a prototype tumor suppressor and an enigma. Genes Dev 30, 1492–1502 (2016). 213. Lohmann, D. R. RB1 gene mutations in retinoblastoma. Hum. Mutat. 14, 283–288 (1999). 214. Herschkowitz, J. I., He, X., Fan, C. & Perou, C. M. The functional loss of the retinoblastoma tumour suppressor is a common event in basal-like and luminal B breast carcinomas. Breast Cancer Res 10, R75 (2008). 215. Cancer Genome Atlas Research Network. Comprehensive genomic characterization of squamous cell lung cancers. Nature 489, 519–525 (2012). 216. Gope, R. et al. Increased expression of the retinoblastoma gene in human colorectal carcinomas relative to normal colonic mucosa. J. Natl. Cancer Inst. 82, 310–314 (1990). 217. Meling, G. I. et al. Genetic alterations within the retinoblastoma in colorectal carcinomas. Relation to DNA ploidy pattern studied by flow cytometric analysis. British Journal of Cancer 64, 475 (1991). 218. Poller, D. N., Baxter, K. J. & Shepherd, N. A. p53 and Rb1 protein expression: are they prognostically useful in colorectal cancer? Br. J. Cancer 75, 87–93 (1997). 219. Ali, A. A. et al. RB1 protein in normal and malignant human colorectal tissue and colon cancer cell lines. FASEB J. 7, 931–937 (1993). 220. Gope, R. & Gope, M. L. Abundance and state of phosphorylation of the retinoblastoma susceptibility gene product in human colon cancer. Mol. Cell. Biochem. 110, 123–133 (1992). 221. Pandey, S., Gordon, P. H. & Wang, E. Expression of proliferation-specific genes in the mucosa adjacent to colon carcinoma. Dis. Colon Rectum 38, 462–467 (1995). 222. Kim, S., Gupta, N. & Pevzner, P. A. Spectral probabilities and generating functions of tandem mass spectra: a strike against decoy databases. J. Proteome Res. 7, 3354–3363 (2008). 223. Kim, S. & Pevzner, P. A. MS-GF+ makes progress towards a universal database search tool for proteomics. Nat Commun 5, 5277 (2014). 224. Beausoleil, S. A., Villén, J., Gerber, S. A., Rush, J. & Gygi, S. P. A probability- based approach for high-throughput protein phosphorylation analysis and site localization. Nat. Biotechnol. 24, 1285–1292 (2006). 225. Forbes, S. A. et al. COSMIC: somatic cancer genetics at high-resolution. Nucleic Acids Res 45, D777–D783 (2017). 226. Jianhong Ou, Yong-Xu Wang & Lihua Julie Zhu. trackViewer: A R/Bioconductor package for drawing elegant interactive tracks or lollipop plot to facilitate integrated analysis of multi-omics data. Bioconductor R package version 1.16.0, (2018). 227. ENCODE Project Consortium. An integrated encyclopedia of DNA elements in the human genome. Nature 489, 57–74 (2012). 228. Liberzon, A. et al. The Molecular Signatures Database (MSigDB) hallmark gene set collection. Cell Syst 1, 417–425 (2015). 229. Yang, W. et al. Genomics of Drug Sensitivity in Cancer (GDSC): a resource for therapeutic biomarker discovery in cancer cells. Nucleic Acids Res. 41, D955-961 (2013).

89 230. Li, J. et al. Characterization of Human Cancer Cell Lines by Reverse-phase Protein Arrays. Cancer Cell 31, 225–239 (2017). 231. Lee, W.-K. et al. Structural and functional insights into the regulation mechanism of CK2 by IP6 and the intrinsically disordered protein Nopp140. Proc. Natl. Acad. Sci. U.S.A. 110, 19360–19365 (2013). 232. Delgado, J. L., Hsieh, C.-M., Chan, N.-L. & Hiasa, H. Topoisomerases as anticancer targets. Biochem. J. 475, 373–398 (2018). 233. Chikamori, K. et al. Phosphorylation of Serine 1106 in the Catalytic Domain of Topoisomerase IIα Regulates Enzymatic Activity and Drug Sensitivity. J. Biol. Chem. 278, 12696–12702 (2003). 234. Kitazawa, H. et al. Ser787 in the proline-rich region of human MAP4 is a critical phosphorylation site that reduces its activity to promote tubulin polymerization. Cell Struct. Funct. 25, 33–39 (2000). 235. Poruchynsky, M. S. et al. Accompanying protein alterations in malignant cells with a microtubule-polymerizing drug-resistance phenotype and a primary resistance mechanism11Abbreviations: MTs, microtubules; MAPs, microtubule-associated proteins; MAP4, microtubule-associated protein-4; PTX, paclitaxel; EPOA, epothilone A; EPOB, epothilone B; EPOA-R, epothilone A-resistant; COL, colchicine; VCR, vincristine; and VBL, vinblastine. Biochemical Pharmacology 62, 1469–1480 (2001). 236. Coss, A. et al. Increased topoisomerase IIalpha expression in colorectal cancer is associated with advanced disease and chemotherapeutic resistance via inhibition of apoptosis. Cancer Lett. 276, 228–238 (2009). 237. de Resende, M. F. et al. Prognostication of prostate cancer based on TOP2A protein and gene assessment: TOP2A in prostate cancer. J Transl Med 11, 36 (2013). 238. Zhang, Y., Baysac, K. C., Yee, L.-F., Saporita, A. J. & Weber, J. D. Elevated DDX21 regulates c-Jun activity and rRNA processing in human breast cancers. Breast Cancer Res. 16, 449 (2014). 239. LI, X.-P. et al. Overexpression of ribosomal L1 domain containing 1 is associated with an aggressive phenotype and a poor prognosis in patients with prostate cancer. Oncol Lett 11, 2839–2844 (2016). 240. Jaiswal, J. K. et al. S100A11 is required for efficient plasma membrane repair and survival of invasive cancer cells. Nature Communications 5, 3795 (2014). 241. De, S., Tsimounis, A., Chen, X. & Rotenberg, S. A. Phosphorylation of α-Tubulin by Stimulates Microtubule Dynamics in Human Breast Cells. Cytoskeleton (Hoboken) 71, 257–272 (2014). 242. Gelais, C. S. et al. A Putative Cyclin-binding Motif in Human SAMHD1 Contributes to Protein Phosphorylation, Localization, and Stability. J. Biol. Chem. 291, 26332– 26342 (2016). 243. Yang, J.-J. et al. ZAK inhibits human lung cancer cell growth via ERK and JNK activation in an AP-1-dependent manner. Cancer Sci. 101, 1374–1381 (2010).

90 244. Yamamoto, H. et al. Paradoxical increase in retinoblastoma protein in colorectal carcinomas may protect cells from apoptosis. Clin. Cancer Res. 5, 1805–1815 (1999). 245. Indovina, P., Pentimalli, F., Casini, N., Vocca, I. & Giordano, A. RB1 dual role in proliferation and apoptosis: Cell fate control and implications for cancer therapy. Oncotarget 6, 17873–17890 (2015). 246. Arhel, N. J. et al. The retinoblastoma protein interacts with Bag-1 in human colonic adenoma and carcinoma derived cell lines. Int. J. Cancer 106, 364–371 (2003). 247. Bullock, N. & Oltean, S. The many faces of SRPK1. J Pathol 241, 437–440 (2017). 248. Shi, L., Pan, H., Liu, Z., Xie, J. & Han, W. Roles of PFKFB3 in cancer. Signal Transduction and Targeted Therapy 2, 17044 (2017). 249. Giuliano, C. J., Lin, A., Smith, J. C., Palladino, A. C. & Sheltzer, J. M. MELK expression correlates with tumor mitotic activity but is not required for cancer growth. eLife 7, 250. Xie, X. et al. A comparative phosphoproteomic analysis of a human tumor metastasis model using a label-free quantitative approach. Electrophoresis 31, 1842–1852 (2010). 251. Gannon, J., Staunton, L., O’Connell, K., Doran, P. & Ohlendieck, K. Phosphoproteomic analysis of aged skeletal muscle. Int. J. Mol. Med. 22, 33–42 (2008). 252. Zimman, A. et al. Activation of aortic endothelial cells by oxidized phospholipids: a phosphoproteomic analysis. J Proteome Res 9, 2812–2824 (2010). 253. Klaeger, S. et al. The target landscape of clinical kinase drugs. Science 358, eaan4368 (2017). 254. Mukherjee, S. et al. Yersinia YopJ Acetylates and Inhibits Kinase Activation by Blocking Phosphorylation. Science 312, 1211–1214 (2006).

91